Anytype vs Poly

Agent access

1. ai-native userPoint an agent at llms.txt or agent-oriented docs

   weight 2 · round to Poly
   Anytype✓fullprobed8/10

   A direct probe confirms Anytype hosts an llms.txt file at doc.anytype.io/llms.txt returning HTTP 200 with organized links to agent-oriented docs, and the developer docs explicitly provide an 'anytype-agents-skill' page meant to be dropped into Claude Code, Cursor, Gemini CLI, etc. so agents can operate on Anytype data directly. This shows both a machine-readable llms.txt entrypoint and human-authored agent-oriented documentation designed for pointing coding agents at. missing for 10: no independent/community confirmation that agents actually consume llms.txt successfully in practice beyond the vendor's own docs and probe.

   • [probe] “PROBE llms.txt: HTTP 200 at https://doc.anytype.io/llms.txt # Anytype Docs ## English - [Welcome](https://doc.anytype.io/anytype/getting-s…”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “an open-source toolkit that gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”

   proof ↗⚑ flag

   Poly✓fullprobed9/10

   Poly directly serves an llms.txt with a full agent-oriented documentation index, confirmed live via runtime probe (HTTP 200, structured docs list, companion llms-full.txt, /raw/*.md mirrors of every page), explicitly designed for agent consumption alongside CLI and MCP integration docs. Missing for 10: no independent third-party report of an agent actually using llms.txt to onboard, only first-party docs and probes.

   • [probe] “PROBE llms.txt: HTTP 200 at https://docs.poly.app/llms.txt # docs.poly.app > Poly documentation ## Documentation Sets - [docs.poly.app](h…”
   • [probe] “PROBE runtime (recorded 2026-09-14): https://docs.poly.app/llms.txt served an agent-oriented docs index ('# docs.poly.app', with a companion…”
   • [claimed-docs] “Import your existing files from Google Drive, Dropbox, iCloud, or your local machine.”

   proof ↗⚑ flag

2. ai-native userRun the product headlessly / in CI for automation

   weight 2 · round to Anytype
   Anytype~partialprobed5/10

   Anytype's docs explicitly claim a CLI that can run Anytype as a headless server for automation and scripting (anytype-docs-2), and a Docker image is maintained for self-hosted network use (anytype-docs-19). However, the only runtime evidence available is for the MCP integration, which requires the desktop GUI app to be actively running and reachable ('Can't connect to API... ensure Anytype is running'), undercutting a fully headless story for the primary automation surface, and there is no independent corroboration of the CLI's headless-server behavior itself. Missing for 10: hands-on verification that the CLI truly runs without any GUI/app dependency, CI-pipeline examples, and independent reports of headless automation in production.

   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”
   • [claimed-docs] “There's also a docker image that's maintained by the team.”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”

   proof ↗⚑ flag

   Poly~partialprobed4/10

   The `poly` CLI is explicitly described as 'scriptable from the terminal' (poly-docs-17) and supports commands like `poly search` (poly-docs-6), suggesting scriptable automation is possible, and the MCP server exists for programmatic agent access (poly-probe-rt-2). However, both the CLI and MCP server rely on browser-based OAuth authentication (poly-probe-rt-2), which is a significant barrier to headless/CI use, and there is no documentation of API keys, service accounts, or CI-specific setup for non-interactive automation. Missing for 10: documented headless/non-interactive auth method (API key or service token), explicit CI/automation examples or docs, evidence of running without a browser session.

   • [claimed-docs] “`poly` is Poly's command-line tool: your files, searchable and scriptable from the terminal.”
   • [claimed-docs] “poly search "the contract with the March deadline"”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”
   • [claimed-docs] “Connect Poly to AI assistants and code editors with the local CLI first, then set up an MCP integration.”

   proof ↗⚑ flag

3. ai-native userPlug MCP servers into this product so it can use their tools

   weight 3 · round drawn
   Anytype—none0/10

   All evidence describes Anytype exposing its own local API as an MCP server for external AI assistants (Claude, Cursor, etc.) to consume — the reverse direction of this story, which asks whether Anytype itself can plug in and use external MCP servers' tools. No evidence shows Anytype acting as an MCP client that consumes third-party MCP servers.

   ⚑ flag

   Poly—none0/10

   All MCP-related evidence describes Poly exposing its own hosted MCP server (mcp.poly.app) for external AI assistants like Claude Code, Codex, ChatGPT, etc. to consume — i.e., Poly acting as the MCP *server* — not Poly acting as an MCP *client* that can plug in and use tools from third-party MCP servers. No documentation, CLI reference, or probe shows Poly's built-in agent connecting to or consuming external MCP servers.

   • [claimed-docs] “The Model Context Protocol (MCP) lets compatible AI assistants connect to Poly.”
   • [claimed-docs] “Connect Poly to AI assistants and code editors with the local CLI first, then set up an MCP integration.”
   • [claimed-docs] “The official plugin loads your Poly memory when each session starts, teaches Claude how to use Poly, and connects the Poly MCP server.”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [probe] “official MCP server documented at https://docs.poly.app/integrations/using-mcp”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

4. ai-native userConnect an agent via an official MCP server

   weight 3 · round to Poly
   Anytype✓fullprobed8/10

   Anytype ships an official MCP server (npm @anyproto/anytype-mcp) documented at developers.anytype.io, confirmed via runtime probe to launch and connect to the local Anytype API, enabling AI assistants to read/search/modify Objects through natural language. missing for 10: independent third-party review of the MCP server in production use beyond the vendor's own probe/docs.

   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language”
   • [probe] “official MCP server documented at https://developers.anytype.io/docs/examples/featured/mcp/”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [claimed-docs] “gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “an open-source toolkit that gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”

   proof ↗⚑ flag

   Poly✓fullprobed9/10

   Poly documents and runtime-verifies an official hosted MCP server (mcp.poly.app/mcp) with OAuth auth, exposing search/fetch/CLI tools, and it's usable from Claude Code, ChatGPT, Codex, Gemini, and other MCP clients. missing for 10: independent third-party (non-vendor) confirmation of a successful end-to-end agent connection beyond the probe and docs.

   • [claimed-docs] “The Model Context Protocol (MCP) lets compatible AI assistants connect to Poly.”
   • [claimed-docs] “Connect Poly to AI assistants and code editors with the local CLI first, then set up an MCP integration.”
   • [claimed-docs] “The official plugin loads your Poly memory when each session starts, teaches Claude how to use Poly, and connects the Poly MCP server.”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [probe] “official MCP server documented at https://docs.poly.app/integrations/using-mcp”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

5. ai-native userUse an official CLI

   weight 2 · round to Poly
   Anytype✓fullclaimed7/10

   Anytype explicitly ships an official CLI ('Anytype CLI is now available - run Anytype as a headless server for automation and scripting'), positioned for scripting/automation use cases relevant to AI-native/agentic workflows. However, evidence is limited to a single doc mention with no independent hands-on verification of the CLI itself (as opposed to the API/MCP server, which is separately corroborated). Missing for 10: independent/community confirmation of the CLI in use, detailed CLI command reference or examples.

   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”

   proof ↗⚑ flag

   Poly✓fullprobed8/10

   Poly ships a documented official CLI (`poly`) with 25+ commands for searching, reading, creating, and organizing files from the terminal, plus an agent-skill that composes CLI commands, confirmed by both docs and a runtime probe. missing for 10: independent/hands-on community verification of the CLI itself (community evidence discusses filesystem access concerns, not CLI usage) and no visible changelog/versioning depth for the CLI beyond docs.

   • [claimed-docs] “poly search "the contract with the March deadline"”
   • [claimed-docs] “`poly` is Poly's command-line tool: your files, searchable and scriptable from the terminal.”
   • [claimed-docs] “The [`poly-cli` agent skill] teaches an agent how to search, read, create, update, and organize your Poly files by composing `poly` commands…”
   • [probe] “official CLI documented at https://docs.poly.app/cli”
   • [probe] “PROBE runtime (recorded 2026-09-14): https://docs.poly.app/llms.txt served an agent-oriented docs index ('# docs.poly.app', with a companion…”

   proof ↗⚑ flag

6. ai-native userDrive the product through a documented public API

   weight 3 · round to Anytype
   Anytype✓fullprobed8/10

   Anytype ships a documented local REST API (OpenAPI 3.1.0 spec, 22 paths, generated developer reference) plus an official MCP server package that fronts it, and a CLI/headless server mode for automation — all confirmed by both docs and runtime probes. missing for 10: independent third-party developer testimonials building against the API beyond the vendor's own probes, and no public hosted/remote API (localhost-only).

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [github] “Extensible through a gRPC API and AI "Agents" (see AGENTS.md).”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”

   proof ↗⚑ flag

   Poly~partialprobed7/10

   Poly documents a CLI (`poly` with 25+ commands) and a hosted MCP server (https://mcp.poly.app/mcp with OAuth) that let AI agents and assistants drive file search, read, write, and organization programmatically, confirmed by both docs and a live runtime probe. However, there is no conventional public REST/OpenAPI API — probes for openapi.json/swagger.json all returned 404 — so agentic access is limited to MCP/CLI surfaces rather than a general-purpose documented API. missing for 10: a REST/OpenAPI-documented API, independent third-party corroboration of API usage beyond vendor docs.

   • [claimed-docs] “`poly` is Poly's command-line tool: your files, searchable and scriptable from the terminal.”
   • [claimed-docs] “poly search "the contract with the March deadline"”
   • [claimed-docs] “Connect Poly to AI assistants and code editors with the local CLI first, then set up an MCP integration.”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.poly.app/openapi.json, https://docs.poly.app/swagger.json, https://docs.poly.app/api/op…”
   • [probe] “official MCP server documented at https://docs.poly.app/integrations/using-mcp”
   • [probe] “official CLI documented at https://docs.poly.app/cli”
   • [probe] “PROBE runtime (recorded 2026-09-14): https://docs.poly.app/llms.txt served an agent-oriented docs index ('# docs.poly.app', with a companion…”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

7. ai-native userIssue scoped/least-privilege API credentials for an agent

   weight 2 · round drawn
   Anytype—none0/10

   Anytype's local API and MCP server require a key minted in the desktop app (anytype-probe-rt-1, anytype-docs-1), but nothing in the evidence describes scopes, permission levels, or least-privilege credential issuance for agents — the API/OpenAPI spec is just a flat set of endpoints (anytype-probe-rt-2) with no mention of restricting an agent's key to specific actions or Objects. missing for 10: any documentation of scoped/role-based API keys, per-agent permission restriction, or revocation/least-privilege credential model.

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's MCP/agent integrations use a single OAuth login granting 'full access to your Poly files' (poly-docs-30, poly-docs-29) rather than any scoped or least-privilege credential/token mechanism; no docs mention granular permission scopes, read-only vs write tokens, or per-agent credential issuance. This axis is applicable (Poly ships agent/API integrations) but no evidence supports scoped credentialing — missing for 10: any documented scope/permission model, token-level restriction, or least-privilege API key issuance for agents.

   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [claimed-docs] “Give Claude Code persistent context and full access to your Poly files.”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

8. ai-native userBuild against official SDKs

   weight 2 · round to Anytype
   Anytype✓fullprobed8/10

   Anytype ships an official local REST API (with a machine-readable OpenAPI 3.1.0 spec covering 22 paths) built into the desktop app, a CLI, and an official MCP server (@anyproto/anytype-mcp) verified at runtime to actually connect and mint keys from the app — giving AI-native developers multiple official, documented SDK-like surfaces. Missing for 10: independent third-party SDK wrappers or client libraries beyond the official API/CLI/MCP, and broader language-specific SDK docs (only OpenAPI + CLI + MCP confirmed).

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [probe] “official MCP server documented at https://developers.anytype.io/docs/examples/featured/mcp/”

   proof ↗⚑ flag

   Poly—none0/10

   Evidence shows a CLI and an MCP server integration, but no official SDK (Python/JS/etc.) or API client library is documented; the OpenAPI probe returned 404s across all candidate paths, indicating no formal API spec developers could build an SDK against.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.poly.app/openapi.json, https://docs.poly.app/swagger.json, https://docs.poly.app/api/op…”
   • [claimed-docs] “`poly` is Poly's command-line tool: your files, searchable and scriptable from the terminal.”
   • [probe] “official MCP server documented at https://docs.poly.app/integrations/using-mcp”

   proof ↗⚑ flag

9. ai-native userSubscribe to events via webhooks

   weight 2 · round drawn
   Anytype—none0/10

   No evidence of webhook subscription support anywhere in Anytype's API/local-API docs; the local API is a localhost REST/gRPC surface with an MCP server for AI assistant access, but nothing indicates event-driven webhook callbacks. Missing for 10: any documentation of webhook registration, event subscription endpoints, or push-notification mechanism.

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [probe] “PROBE openapi: all candidate paths 404 (https://doc.anytype.io/openapi.json, https://doc.anytype.io/swagger.json, https://doc.anytype.io/api…”

   proof ↗⚑ flag

   Poly—none0/10

   No evidence of any webhook or event-subscription mechanism; Poly's integration surface is limited to MCP, CLI, and file sync, none of which describes outbound event notifications via webhooks.

   ⚑ flag

Agentic features

1. ai-native userGet AI-generated insights and suggestions from my data inside the product

   weight 2 · round to Poly
   Anytype~partialprobed5/10

   Anytype doesn't ship a native in-app AI-insights feature, but its local API + official MCP server + 'Agents Skill' let external AI assistants (Claude, Cursor, Gemini CLI, etc.) read, search, and run scripts against a user's Anytype data to generate insights/suggestions, and Queries provide auto-updating views of the knowledge base. This is enablement via integration rather than a built-in 'AI-generated insights' feature inside the product itself. Missing for 10: no evidence of a native, first-party AI panel/chat that autonomously surfaces insights without wiring up an external agent, and no independent hands-on account of the agent-generated insights actually working well.

   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “an open-source toolkit that gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [claimed-docs] “A Query is a question you ask your knowledge base—answered instantly and kept up to date automatically.”
   • [claimed-docs] “Rather than manually curating a list of things you're looking for, you define the criteria and the Query assembles the result for you as a V…”

   proof ↗⚑ flag

   Poly✓fullprobed7/10

   Poly's docs describe natural-language search, an automatic 'second brain' memory layer, and an agent that can summarize, explain, and generate insights from indexed files (docs-2, docs-3, docs-31, docs-14), directly matching the AI-generated insights/suggestions story. Missing for 10: independent/hands-on evidence of proactive insight generation (vs. just search/summarize-on-request) and no third-party validation of insight quality beyond vendor docs.

   • [claimed-docs] “Search inside your files with natural language that's gives you results down to the exact timestamp, page, paragraph, or frame.”
   • [claimed-docs] “Poly automatically remembers key details, forming a "second brain" for you out of your files.”
   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “Ask it to summarize, explain, or even edit your files. Automate tasks like organizing a folder.”
   • [claimed-docs] “It automatically parses, transcribes, and indexes your files that make it up to 20x faster for an agent to read and understand them. Use our…”
   • [probe] “PROBE runtime (recorded 2026-09-14): https://docs.poly.app/llms.txt served an agent-oriented docs index ('# docs.poly.app', with a companion…”

   proof ↗⚑ flag

2. ai-native userSet up automations that run autonomously in the background

   weight 2 · round to Poly
   Anytype~partialprobed4/10

   Anytype exposes a local API, CLI (headless server), and an official MCP server plus an 'agents-skill' that lets external AI agents (Claude Code, Cursor, etc.) run scripts against Anytype data, giving building blocks for automation. However there is no documented scheduler, trigger, or persistent background-running agent inside Anytype itself — all examples describe an external agent tool invoked on-demand, not autonomous background operation. Missing for 10: evidence of a built-in scheduler/trigger system, proof that agents can run unattended over time rather than per-invocation, and independent confirmation of autonomous background execution.

   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”
   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “official MCP server documented at https://developers.anytype.io/docs/examples/featured/mcp/”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”

   proof ↗⚑ flag

   Poly~partialprobed5/10

   Poly's agent can perform multi-step file tasks, spawn subagents, and be invoked via CLI/MCP/plugins, but these are triggered interactively (chat requests, CLI commands, agent sessions) rather than described as scheduled or event-triggered background automations running autonomously without a user prompt. There's no documented scheduler, trigger system, or 'run this automatically every day/on file change' capability. missing for 10: explicit scheduling/trigger mechanism (cron-like, event-driven, or always-on background execution), evidence of automations persisting and running unattended over time, and independent confirmation of autonomous background operation.

   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “Your agent can spawn its own subagents, assign them tasks, and track their progress.”
   • [claimed-docs] “Your agent can spawn its own subagents, assign them tasks, and track their progress. Moreover, agents/subagents can communicate with each ot…”
   • [claimed-docs] “Ask it to summarize, explain, or even edit your files. Automate tasks like organizing a folder.”
   • [claimed-docs] “The official plugin loads your Poly memory when each session starts, teaches Claude how to use Poly, and connects the Poly MCP server.”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

3. ai-native userDelegate tasks to a built-in AI assistant inside the product

   weight 3 · round to Poly
   Anytype—none0/10

   Evidence shows Anytype exposes a local API, CLI, and MCP server that let *external* AI assistants (Claude Code, Cursor, Gemini CLI, etc.) read/write Anytype data — this is Anytype acting as a data backend for outside agents, not a built-in AI assistant living inside the product that a user can delegate tasks to. No docs mention an in-app chat/assistant feature.

   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [github] “Extensible through a gRPC API and AI "Agents" (see AGENTS.md).”

   proof ↗⚑ flag

   Poly✓fullprobed8/10

   Poly ships a built-in agent that can search, read, write, edit, and organize files across multiple steps (poly-docs-14, poly-docs-27, poly-docs-31), and can even spawn subagents to delegate subtasks (poly-docs-21, poly-docs-35), directly matching the story of delegating tasks to an in-product AI assistant. This is corroborated by runtime probes confirming a live agent/MCP infrastructure (poly-probe-rt-1, poly-probe-rt-2). Missing for 10: independent hands-on user reports specifically validating the agent's task execution (community commentary focuses on search/security concerns rather than confirming agentic task delegation in practice).

   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “The Poly agent can write and edit files directly in your storage.”
   • [claimed-docs] “Ask it to summarize, explain, or even edit your files. Automate tasks like organizing a folder.”
   • [claimed-docs] “Your agent can spawn its own subagents, assign them tasks, and track their progress.”
   • [claimed-docs] “Your agent can spawn its own subagents, assign them tasks, and track their progress. Moreover, agents/subagents can communicate with each ot…”
   • [probe] “PROBE runtime (recorded 2026-09-14): https://docs.poly.app/llms.txt served an agent-oriented docs index ('# docs.poly.app', with a companion…”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

4. ai-native userOperate the product with natural-language commands

   weight 2 · round drawn
   Anytype✓fullprobed8/10

   Anytype ships an official MCP server (npm @anyproto/anytype-mcp) that fronts its local API, explicitly documented as 'enabling AI assistants to seamlessly interact with Anytype's API through natural language,' verified working at runtime against the local app. Additionally an 'agents skill' lets agent-capable tools (Claude Code, Cursor, Gemini CLI, Copilot) run natural-language-driven scripts against Anytype data. missing for 10: independent third-party review of natural-language command accuracy/robustness beyond vendor docs and a single runtime probe.

   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language”
   • [claimed-docs] “an open-source toolkit that gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “official MCP server documented at https://developers.anytype.io/docs/examples/featured/mcp/”

   proof ↗⚑ flag

   Poly✓fullprobed8/10

   Poly's docs describe a built-in agent operable via natural language — searching, summarizing, editing, and organizing files with plain-language requests (poly-docs-2, poly-docs-31, poly-docs-14, poly-docs-27), plus a CLI and MCP-based integrations exposing the same commands to AI assistants. This directly matches the ai-native natural-language-command story. Missing for 10: independent/hands-on verification that NL commands work reliably in practice (community evidence only discusses trust/security concerns, not functional testing of NL command execution).

   • [claimed-docs] “Search inside your files with natural language that's gives you results down to the exact timestamp, page, paragraph, or frame.”
   • [claimed-docs] “poly search "the contract with the March deadline"”
   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “The Poly agent can write and edit files directly in your storage.”
   • [claimed-docs] “Ask it to summarize, explain, or even edit your files. Automate tasks like organizing a folder.”
   • [probe] “PROBE runtime (recorded 2026-09-14): https://docs.poly.app/llms.txt served an agent-oriented docs index ('# docs.poly.app', with a companion…”

   proof ↗⚑ flag

Api quality

1. ai-native userExplore an interactive API reference with runnable examples

   weight 2 · round to Anytype
   Anytype~partialprobed5/10

   Anytype ships a genuine local REST API with a machine-readable OpenAPI 3.1 spec (22 paths) and a developer reference site (developers.anytype.io) generated from it, which documents endpoints like chat/messages with version tags — this is the backbone of an API reference. However, there's no direct evidence of an interactive 'try it out' console or runnable code examples embedded in the docs; probes only confirm the spec's existence and doc generation, not interactivity. Missing for 10: explicit evidence of a live 'try it' widget or runnable code snippets, independent confirmation of interactive testing within the docs UI.

   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [claimed-docs] “Create chats, send and stream messages, react, and search chat history (anytype-heart v0.50.7)”
   • [probe] “PROBE openapi: all candidate paths 404 (https://doc.anytype.io/openapi.json, https://doc.anytype.io/swagger.json, https://doc.anytype.io/api…”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”

   proof ↗⚑ flag

   Poly—none0/10

   The probe explicitly found no OpenAPI/Swagger spec at any candidate path, and the evidence pack contains no mention of an interactive API reference or runnable examples anywhere in Poly's docs or CLI/MCP materials.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.poly.app/openapi.json, https://docs.poly.app/swagger.json, https://docs.poly.app/api/op…”

   proof ↗⚑ flag

2. ai-native userDownload a machine-readable API spec (OpenAPI or equivalent)

   weight 2 · round to Anytype
   Anytype✓fullprobed9/10

   A genuine machine-readable OpenAPI 3.1.0 spec (22 paths) for Anytype's local API is confirmed to exist and was fetched directly from the vendor's core repo, and developers.anytype.io's API reference is generated from this same spec — i.e. a real downloadable API spec exists and is used to power official docs. Missing for 10: the spec isn't hosted at a stable public URL on the main docs domain (doc.anytype.io openapi.json paths returned 404), so discovery requires knowing the GitHub repo path rather than a standard /openapi.json convention.

   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [probe] “PROBE openapi: all candidate paths 404 (https://doc.anytype.io/openapi.json, https://doc.anytype.io/swagger.json, https://doc.anytype.io/api…”
   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [claimed-docs] “Create chats, send and stream messages, react, and search chat history (anytype-heart v0.50.7)”

   proof ↗⚑ flag

   Poly—none0/10

   A direct probe for OpenAPI/Swagger specs at all standard paths returned 404s, and no evidence pack item mentions a downloadable API spec; Poly instead documents an MCP server and CLI, not a machine-readable REST API spec.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.poly.app/openapi.json, https://docs.poly.app/swagger.json, https://docs.poly.app/api/op…”

   proof ↗⚑ flag

3. ai-native userTest against a sandbox environment without touching production data

   weight 1 · round drawn
   Anytype—none0/10

   Anytype's docs describe local-only mode, self-hosting, and a local API/MCP server that operates directly on the user's real Anytype data, but there is no mention of a dedicated sandbox/test space, staging environment, or mock-data mode that lets an AI agent test actions without touching production content. Evidence shows agents run scripts 'against your Anytype data' directly, with no isolation layer.

   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [claimed-docs] “If you prefer to fully keep your data on your device without any server to sync between your desktop and mobile, you can enable local-only m…”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's docs describe direct agent read/write access to a user's real files, Drive/Dropbox imports, and MCP integrations, but nowhere mention a sandbox, staging, or test environment isolated from production data. Community comments even highlight anxiety about the agent having broad access to real files with no isolation layer. missing for 10: any documented sandbox/staging mode, test data set, or isolated environment separate from the user's live files.

   • [claimed-docs] “The Poly agent can write and edit files directly in your storage.”
   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [community] “Hooking up the Internet to my filesystem is scary. What security measures are in place to ensure a compromise of your infrastructure doesn't…”
   • [community] “With Google Drive, I choose which files to upload. It doesn't have broad access to everything on my computer... your service, by nature, is …”

   proof ↗⚑ flag

4. ai-native userRely on versioned APIs with a documented deprecation policy

   weight 2 · round drawn
   Anytype—none0/10

   Evidence confirms Anytype ships a local API with an OpenAPI spec that is date-versioned (2025-04-22), but nowhere in the docs or probes is there any documented deprecation policy, versioning guarantees, or migration/sunset process for API changes. Absence of evidence for this applicable axis (a real, documented, self-hosted API exists) means the story is not delivered.

   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [probe] “PROBE openapi: all candidate paths 404 (https://doc.anytype.io/openapi.json, https://doc.anytype.io/swagger.json, https://doc.anytype.io/api…”
   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”

   proof ↗⚑ flag

   Poly—none0/10

   No evidence of any versioned public API or documented deprecation policy — probes for OpenAPI/Swagger specs all returned 404, and docs only cover CLI/MCP/agent integrations, not API versioning or deprecation guarantees. missing for 10: versioned API reference, deprecation policy documentation, changelog/versioning scheme.

   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.poly.app/openapi.json, https://docs.poly.app/swagger.json, https://docs.poly.app/api/op…”

   proof ↗⚑ flag

Agent ops

1. ai-native userMy coding agent or assistant can read AND write my notes headlessly — through a documented local API, vendor API, or MCP server — without me touching the GUI

   weight 3 · round drawn
   Anytype✓fullprobed9/10

   Anytype documents a local API and an official MCP server (@anyproto/anytype-mcp) that fronts that API, allowing AI assistants like Claude Code, Cursor, etc. to read, search, and modify Objects entirely headlessly without touching the GUI, and this was verified at runtime to actually launch and connect. The OpenAPI spec (22 paths) confirms concrete read/write endpoints beyond marketing claims. Missing for 10: independent third-party (non-vendor) hands-on report confirming write operations succeed end-to-end via the MCP server.

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “an open-source toolkit that gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “official MCP server documented at https://developers.anytype.io/docs/examples/featured/mcp/”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”

   proof ↗⚑ flag

   Poly✓fullprobed9/10

   Poly ships a documented hosted MCP server (mcp.poly.app) with OAuth, exposing search, fetch, and CLI-equivalent tools, plus a CLI/agent-skill that lets agents read, write, create, and organize files headlessly, and a Claude Code plugin that loads Poly memory and connects the MCP server automatically — all confirmed live via runtime probes. missing for 10: independent (non-vendor) hands-on confirmation that write operations via MCP work reliably in practice, beyond docs and probes.

   • [claimed-docs] “The Model Context Protocol (MCP) lets compatible AI assistants connect to Poly.”
   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “The official plugin loads your Poly memory when each session starts, teaches Claude how to use Poly, and connects the Poly MCP server.”
   • [claimed-docs] “The [`poly-cli` agent skill] teaches an agent how to search, read, create, update, and organize your Poly files by composing `poly` commands…”
   • [claimed-docs] “The Poly agent can write and edit files directly in your storage.”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [probe] “official MCP server documented at https://docs.poly.app/integrations/using-mcp”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

Ai assist

1. ai-native userConnect external LLMs and AI tools to my notes through first-party connectors or an ecosystem of AI plugins

   weight 2 · round to Poly
   Anytype✓fullprobed8/10

   Anytype ships a first-party local API plus an official MCP server (npm @anyproto/anytype-mcp) confirmed working at runtime, enabling AI assistants like Claude, Cursor, Gemini CLI, GitHub Copilot to read/search/modify Objects, alongside an 'Agents Skill' toolkit for agent-capable tools. missing for 10: no broader third-party AI plugin marketplace/ecosystem beyond the official MCP server and agents-skill, and no independent community review of the MCP integration's real-world reliability.

   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language”
   • [claimed-docs] “gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “an open-source toolkit that gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [github] “Extensible through a gRPC API and AI "Agents" (see AGENTS.md).”

   proof ↗⚑ flag

   Poly✓fullprobed9/10

   Poly ships a first-party MCP server (mcp.poly.app) with OAuth, documented and runtime-verified as live, plus official plugins/integrations for Claude, ChatGPT, Claude Code, VS Code, and a CLI usable by agents, giving AI-native users multiple first-party ways to connect external LLMs to their notes/files. missing for 10: no independent hands-on confirmation of the plugin ecosystem working smoothly beyond docs/probe, and community feedback raises unresolved trust/security concerns rather than functional failures.

   • [claimed-docs] “The Model Context Protocol (MCP) lets compatible AI assistants connect to Poly.”
   • [claimed-docs] “Connect Poly to AI assistants and code editors with the local CLI first, then set up an MCP integration.”
   • [claimed-docs] “The official plugin loads your Poly memory when each session starts, teaches Claude how to use Poly, and connects the Poly MCP server.”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [claimed-docs] “Give Claude Code persistent context and full access to your Poly files.”
   • [probe] “official MCP server documented at https://docs.poly.app/integrations/using-mcp”
   • [probe] “PROBE runtime (recorded 2026-09-14): https://docs.poly.app/llms.txt served an agent-oriented docs index ('# docs.poly.app', with a companion…”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

2. ai-native userA built-in AI assistant can answer questions, summarize, and draft grounded in my own notes — with clear controls over what data it sees

   weight 2 · round to Poly
   Anytype—none0/10

   Anytype ships a local API and an MCP server that let external AI assistants (Claude, Cursor, etc.) read/search/modify Objects, but there is no evidence of a built-in in-app AI assistant that itself answers questions, summarizes, or drafts grounded in the user's notes with UI-level data-scope controls — the docs describe it as a toolkit for external agents, not a native assistant feature. Missing for 10 (or even partial credit): a first-party in-app assistant UI, summarization/drafting features, and explicit data-scope/consent controls within Anytype itself.

   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language”
   • [claimed-docs] “gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “an open-source toolkit that gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”

   proof ↗⚑ flag

   Poly~partialcommunity5/10

   Poly's built-in agent clearly summarizes, explains, drafts, and edits notes grounded in the user's own files (poly-docs-31, poly-docs-14, poly-docs-27, poly-docs-3), and Mind/journaling gives a personal-notes corpus for it to draw on (poly-docs-12, poly-docs-26, poly-docs-32). However, evidence for granular controls over *what the assistant itself sees* is thin: docs describe the agent getting 'full access to your Poly files' (poly-docs-30) rather than scoped visibility, and the only access controls documented are sharing/revoke permissions for other people (poly-docs-9, poly-docs-33), not data-scoping for the AI itself — a gap community users flag as a trust concern (poly-comm-2, poly-comm-4). Missing for 10: explicit per-folder/file visibility toggles or exclusion settings for the assistant, and independent confirmation that such controls work as intended.

   • [claimed-docs] “Ask it to summarize, explain, or even edit your files. Automate tasks like organizing a folder.”
   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “The Poly agent can write and edit files directly in your storage.”
   • [claimed-docs] “Poly automatically remembers key details, forming a "second brain" for you out of your files.”
   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “Give Claude Code persistent context and full access to your Poly files.”
   • [claimed-docs] “Share a file or folder when the content belongs to you and you want specific people — or anyone with a link — to see it.”
   • [claimed-docs] “Share a file or folder when the content belongs to you and you want specific people — or anyone with a link — to see it. You stay the owner,…”
   • [community] “Hooking up the Internet to my filesystem is scary. What security measures are in place to ensure a compromise of your infrastructure doesn't…”
   • [community] “With Google Drive, I choose which files to upload. It doesn't have broad access to everything on my computer... your service, by nature, is …”

   proof ↗⚑ flag

1. ai-native userPerform bulk operations across many items at once

   weight 2 · round to Anytype
   Anytype~partialprobed6/10

   Anytype documents an agent skill where an AI agent can run scripts against Anytype data to handle bulk edits transformations and repetitive tasks using the local API and MCP server plus a CLI for headless automation. However this relies entirely on external scripting and agent tooling rather than a native bulk edit UI and a community report flags friction with bulk CSV merges into databases. missing for 10: native in-app bulk edit selection tools, independent hands-on verification of large-scale bulk script runs, resolution of the community reported CSV merge gap

   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [claimed-docs] “gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [community] “I consistently find myself in need of a 'merge feature' when re-importing cleaned CSV data into Anytype database objects - the larger the da…”

   proof ↗⚑ flag

   Poly~partialclaimed5/10

   Poly's CLI is described as 'scriptable from the terminal' (poly-docs-17) and the agent can chain multi-step tasks like 'organizing a folder' or spawning subagents to handle work in parallel (poly-docs-14, poly-docs-21, poly-docs-31, poly-docs-35), which implies some automation across sets of files, but there is no explicit documentation of a true bulk/batch operation (e.g., 'rename/tag/move N files at once') or a dedicated batch API/command. Missing for 10: an explicit bulk-operation example or batch command, evidence of scale (many items processed in one call), and independent/hands-on confirmation that automation reliably spans large item sets.

   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “`poly` is Poly's command-line tool: your files, searchable and scriptable from the terminal.”
   • [claimed-docs] “Your agent can spawn its own subagents, assign them tasks, and track their progress.”
   • [claimed-docs] “Ask it to summarize, explain, or even edit your files. Automate tasks like organizing a folder.”
   • [claimed-docs] “Your agent can spawn its own subagents, assign them tasks, and track their progress. Moreover, agents/subagents can communicate with each ot…”

   proof ↗⚑ flag

2. ai-native userDefine rules that trigger actions automatically on events

   weight 3 · round drawn
   Anytype—none0/10

   Evidence shows Queries (auto-updating saved views) and an 'Agents skill' letting an external AI agent run scripts against Anytype data on request, plus an MCP/local API for AI assistants to read/write objects — but none of this describes user-defined rules that automatically trigger actions on events (e.g. 'when X happens, do Y'). There is no workflow/automation-rule engine, webhook trigger, or event-based action system documented anywhere in the pack.

   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “A Query is a question you ask your knowledge base—answered instantly and kept up to date automatically.”
   • [claimed-docs] “Rather than manually curating a list of things you're looking for, you define the criteria and the Query assembles the result for you as a V…”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [claimed-docs] “gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's docs describe on-demand agent actions (search, summarize, edit, organize) and multi-step task execution, but there is no evidence of a rules/trigger system that fires actions automatically on defined events (e.g., file-added, schedule, webhook-based automation). Automation is user-initiated per request, not event-driven.

   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “Ask it to summarize, explain, or even edit your files. Automate tasks like organizing a folder.”
   • [claimed-docs] “Your agent can spawn its own subagents, assign them tasks, and track their progress.”

   proof ↗⚑ flag

3. ai-native userSchedule recurring jobs or workflows

   weight 2 · round drawn
   Anytype—none0/10

   Evidence shows Anytype exposes a local API, MCP server, and an 'agents skill' letting external agents run one-off scripts against Anytype data, but nothing describes any built-in scheduler, cron-like trigger, or recurring/automated workflow execution within Anytype itself. Automation-depth is a fair axis for an AI-native knowledge tool, so absence of evidence yields 'none' rather than 'na'.

   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's docs describe agent-driven task automation (multi-step tasks, subagents, organizing folders) but there is no mention of recurring/scheduled execution, cron-like triggers, or workflow scheduling anywhere in the evidence pack.

   ⚑ flag

4. ai-native userVersion, review, and roll back my automations

   weight 1 · round to Poly
   Anytype—none0/10

   Anytype documents AI agent scripting/automation capabilities (agents running scripts against Anytype data) and general object version history is not mentioned anywhere in the evidence, nor is any mechanism to version, review, or roll back automations/scripts themselves. Since Anytype does support automation via its agents-skill and API, this axis is applicable, but no evidence shows versioning, review workflows, or rollback for those automations.

   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [claimed-docs] “gives AI assistants a safe way to read, search, and modify Objects in your Anytype Channels”

   proof ↗⚑ flag

   Poly~partialclaimed4/10

   Poly documents file-level version history and rollback (flashback mode, conflict-preserving recovery) that would cover changes made by its agent when it edits files, but there is no evidence of a first-class 'automation' object (e.g., saved workflows/skills) that can be versioned, reviewed, or rolled back as a unit — only raw file state is versioned. missing for 10: dedicated automation/workflow versioning UI, diff/review of agent actions before execution, rollback of multi-step agent tasks (not just files) as an atomic unit.

   • [claimed-docs] “Even conflicting edits are preserved, so you can recover any version of any file at any time.”
   • [claimed-docs] “Go into flashback mode at any time to see exactly what your files looked like at any point in the past, down to the millisecond.”
   • [claimed-docs] “The Poly agent can handle tasks that span multiple steps — reading, searching, writing, and organizing — in a single request.”
   • [claimed-docs] “The Poly agent can write and edit files directly in your storage.”

   proof ↗⚑ flag

Capture

1. researcherClip web pages, articles, and highlights from my browser straight into my notes with an official clipper or capture flow

   weight 2 · round drawn
   Anytype—none0/10

   No evidence of a browser clipper extension or web-capture flow for Anytype; import features cover Markdown/Any-Block/Obsidian migration, not web page/article clipping. Missing for 10: official browser extension, article/highlight capture flow, bookmarklet or share-to-Anytype integration.

   ⚑ flag

   Poly—none0/10

   Poly's docs describe importing from Drive/Dropbox/iCloud/local files and from note apps like Notion/Obsidian/Evernote, but there is no mention of a browser extension, web clipper, or capture flow for saving web pages/articles/highlights directly into Poly. This is a fair capability for a 'second brain' note tool, so absence of evidence means it is not delivered.

   • [claimed-docs] “Import your existing files from Google Drive, Dropbox, iCloud, or your local machine.”
   • [claimed-docs] “Bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life.”

   proof ↗⚑ flag

Daily notes

1. power-userI get an automatic daily note for each day, pre-filled from a template, as the default landing place for journaling and capture

   weight 3 · round to Poly
   Anytype—none0/10

   Anytype documents general Templates and Types (docs-13, docs-22, docs-32) but nowhere describes an automatic daily/journal note feature that auto-creates a dated entry pre-filled from a template as a default landing page — no evidence of a 'Daily Note' or journaling automation feature exists in the pack.

   • [claimed-docs] “A Template is a saved layout for a page that you can reuse.”
   • [claimed-docs] “you define what a good starting point looks like once and it'll form the base for every new Object you create afterwards”
   • [claimed-docs] “Anytype comes with built-in Types, but you can create your own custom Types to match your needs.”

   proof ↗⚑ flag

   Poly~partialclaimed3/10

   Poly Mind's journaling feature provides a manual 'quiet place' to write daily thoughts and syncs entries as Markdown files, which loosely matches the capture/journaling theme, but there's no evidence of automatic daily note creation, template pre-fill, or it being the default landing screen. missing for 10: automatic daily note generation, template pre-fill mechanism, evidence of default-landing behavior on app open.

   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life. Write a few thoughts, attach files, or add the broader context behind a…”

   proof ↗⚑ flag

Editor

1. power-userDefine reusable templates (with variables like date and title) and apply them when creating recurring kinds of notes

   weight 2 · round to Anytype
   Anytype~partialclaimed6/10

   Anytype's docs clearly describe reusable Templates ('a saved layout for a page that you can reuse' and 'define what a good starting point looks like once... base for every new Object') including customizable cover/icon, which supports recurring note types via Types+Templates. However, there's no evidence of dynamic variable substitution (e.g., auto-filled date, title placeholders) within templates—only static layout/property definitions are documented. missing for 10: explicit variable/placeholder support (auto-date, auto-title fill), independent hands-on confirmation of template variable behavior.

   • [claimed-docs] “A Template is a saved layout for a page that you can reuse.”
   • [claimed-docs] “you define what a good starting point looks like once and it'll form the base for every new Object you create afterwards”
   • [claimed-docs] “Choose an image from the gallery or upload your own image.”
   • [claimed-docs] “Anytype comes with built-in Types, but you can create your own custom Types to match your needs.”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's Mind/journaling docs describe creating markdown journal entries and importing notes, but there is no evidence of a reusable template system with variables (date, title) for recurring note types.

   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life. Write a few thoughts, attach files, or add the broader context behind a…”
   • [claimed-docs] “Bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more.”

   proof ↗⚑ flag

2. power-userWrite and format notes with headings, callouts, tables, code blocks, and embedded media in a fast keyboard-driven editor

   weight 2 · round to Anytype
   Anytype~partialcommunity5/10

   Docs and GitHub confirm a block-based editor (text, databases, kanban, calendar, custom Types) and community reports confirm tables exist (though 'simple, no formulas'), plus templates and page icons/covers. However there is no explicit evidence for headings, callouts, or code blocks as first-class formatting elements, nor any documentation of keyboard shortcuts or 'fast keyboard-driven' editing speed, and embedded media support is only implied via cover images. missing for 10: explicit callout/heading/code-block support, documented keyboard-shortcut system, and evidence of rich embedded media (video/audio/files) in the editor.

   • [github] “Composable blocks: text, databases, kanban, calendar & custom Types.”
   • [community] “Anytype has tables now, simple, no formulas yet. Plan to add formulas to databases - it's one of the top requested features on our forum (co…”
   • [claimed-docs] “A Template is a saved layout for a page that you can reuse.”
   • [claimed-docs] “Anytype comes with built-in Types, but you can create your own custom Types to match your needs.”
   • [community] “I've been using this for about a year. It's very polished and been stable for me. The Apple and Android apps work well too... I'm using it a…”

   proof ↗⚑ flag

   Poly~partialclaimed3/10

   Poly supports Markdown-based journaling (auto-synced .md files in the Mind folder) and offers Vim keybindings for file editing, suggesting some keyboard-driven text editing capability, but there is no documentation of rich formatting elements like callouts, tables, code blocks, or embedded media in a dedicated note editor. Missing for 10: explicit support for headings/callouts/tables/code blocks, embedded media, and any description of a fast dedicated editor UI beyond basic Markdown journaling.

   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “Vim keybindings (enable in Settings)”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life. Write a few thoughts, attach files, or add the broader context behind a…”

   proof ↗⚑ flag

Migration

1. switcherExport my entire knowledge base at any time in open formats, so leaving is as easy as arriving

   weight 2 · round to Anytype
   Anytype✓fullclaimed8/10

   Docs explicitly state you can import content and export it at any time, with Markdown or Any-Block format choices, plus dedicated import from Obsidian and self-host backup/export of channels for migrating identities — covering the core switcher promise of open-format export/import at will. missing for 10: independent hands-on verification of export fidelity/completeness and no explicit mention of bulk/full-space export beyond channel-level export.

   • [claimed-docs] “Choose your preferred format: Markdown or Any-Block.”
   • [claimed-docs] “You can import content into your spaces and export it at any time.”
   • [claimed-docs] “Anytype will process the contents and automatically create a separate object for each Markdown file.”
   • [claimed-docs] “Open Channel Settings → Import → Obsidian and click on Import files.”
   • [claimed-docs] “export all your channels, and import them into a new identity after switching the network mode”
   • [claimed-docs] “To back up your channels and data to a self-hosted network, you need to have a network configuration.”

   proof ↗⚑ flag

   Poly~partialclaimed4/10

   Poly stores files as real files that sync to a local Finder/Explorer folder (poly-docs-23) and journal entries are plain Markdown (poly-docs-12), implying your data always exists in accessible, mostly-original formats rather than a locked proprietary export. However, there is no explicit 'export my knowledge base' feature, bulk-export tool, or open-format guarantee for the indexed metadata/second-brain layer — only import pathways (poly-docs-1, poly-docs-22) are documented. Missing for 10: an explicit export/backup command or API, confirmation that indexed metadata (tags, memory, search index) can be exported, and any open-format guarantee beyond raw file sync.

   • [claimed-docs] “your Poly files appear as a real folder in Finder or Explorer, with files downloading on demand and syncing back automatically.”
   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “Import your existing files from Google Drive, Dropbox, iCloud, or your local machine.”
   • [claimed-docs] “Bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more.”

   proof ↗⚑ flag

2. switcherImport my existing notes from Notion, Evernote, Roam, or plain Markdown with structure, links, and attachments preserved

   weight 3 · round drawn
   Anytype~partialcommunity5/10

   Anytype documents Markdown import (choose Markdown or Any-Block format) and a dedicated Obsidian migration path that creates a separate object per Markdown file, preserving structure to some degree, plus general import/export at any time. However, there is no evidence of dedicated Notion, Evernote, or Roam importers preserving links/attachments as the story specifies — only generic Markdown ingestion is documented, and a community comment even flags gaps in re-importing structured data (CSV) cleanly. Missing for 10: dedicated Notion/Evernote/Roam import tooling, evidence that links and attachments survive import from these specific sources, and independent hands-on confirmation of fidelity.

   • [claimed-docs] “Choose your preferred format: Markdown or Any-Block.”
   • [claimed-docs] “Anytype will process the contents and automatically create a separate object for each Markdown file.”
   • [claimed-docs] “Open Channel Settings → Import → Obsidian and click on Import files.”
   • [claimed-docs] “You can import content into your spaces and export it at any time.”
   • [community] “I consistently find myself in need of a 'merge feature' when re-importing cleaned CSV data into Anytype database objects - the larger the da…”

   proof ↗⚑ flag

   Poly~partialclaimed5/10

   Poly's docs explicitly state you can 'bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more,' directly covering two of the four named sources, and general file import (Google Drive/Dropbox/local) plus native Markdown journal files supports plain Markdown ingestion. However, Roam is never explicitly named, and there is no documented detail on how structure, internal links, or attachments are preserved during the import process, nor any hands-on/community confirmation of import fidelity. Missing for 10: explicit Roam support, documented handling of internal links/structure/attachments during migration, independent corroboration of import fidelity.

   • [claimed-docs] “Bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more.”
   • [claimed-docs] “Import your existing files from Google Drive, Dropbox, iCloud, or your local machine.”
   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”

   proof ↗⚑ flag

Graph

1. researcherExplore my knowledge base as an interactive graph of notes and connections, filtered to what I care about

   weight 1 · round to Anytype
   Anytype✓fullclaimed8/10

   Anytype's Graph is explicitly documented as 'a live, visual map of every Object in your Channel and how they connect to each other,' directly matching the interactive knowledge graph story, and linking is described as core to turning notes into a connected knowledge base. Filtering/querying is supported via the Query feature, which lets users define criteria to assemble filtered Views, though docs don't explicitly confirm Query filters apply directly to the Graph visualization itself. Missing for 10: explicit documentation or hands-on evidence that Graph view can be filtered by Query/View criteria, and independent user corroboration of the graph exploration experience.

   • [claimed-docs] “The Graph is a live, visual map of every Object in your Channel and how they connect to each other.”
   • [claimed-docs] “These connections become incredibly valuable. You can trace how ideas relate, see what's connected to a project”
   • [claimed-docs] “Linking on Anytype is what turns a bunch of separate notes into a connected knowledge base.”
   • [claimed-docs] “A Query is a question you ask your knowledge base—answered instantly and kept up to date automatically.”
   • [claimed-docs] “Rather than manually curating a list of things you're looking for, you define the criteria and the Query assembles the result for you as a V…”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's docs describe a 'second brain' with search, journaling, tagging, multiple folder views (Icon, Grid, Feed, Detail, Tree, etc.), and Notion/Obsidian import, but nowhere mention a graph visualization of notes/connections or filtering such a graph — this is a plausible axis for a notes/second-brain product but no evidence supports it.

   • [claimed-docs] “Poly automatically remembers key details, forming a "second brain" for you out of your files.”
   • [claimed-docs] “Eight ways to look at a folder — Icon, Grid, Feed, Detail, Tree, C”
   • [claimed-docs] “Bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life.”

   proof ↗⚑ flag

Links

1. researcherLink notes with [[wiki-style]] links and see every backlink and unlinked mention of a note in context

   weight 3 · round to Anytype
   Anytype~partialclaimed6/10

   Docs confirm bidirectional-style linking ('Linking on Anytype is what turns a bunch of separate notes into a connected knowledge base', 'trace how ideas relate, see what's connected') and a live Graph view showing every Object's connections, which covers wiki-links and a form of backlink visualization. However, the evidence never explicitly documents a dedicated 'Backlinks' panel or an 'unlinked mentions' detection feature (text mentioning a note's title without a link), so the researcher's specific in-context backlink/unlinked-mention workflow is only partly evidenced. Missing for 10: explicit backlinks list UI, unlinked-mentions detection, independent/hands-on confirmation of these specific sub-features.

   • [claimed-docs] “Linking on Anytype is what turns a bunch of separate notes into a connected knowledge base.”
   • [claimed-docs] “These connections become incredibly valuable. You can trace how ideas relate, see what's connected to a project”
   • [claimed-docs] “The Graph is a live, visual map of every Object in your Channel and how they connect to each other.”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's 'Mind' feature supports journaling, importing notes from other PKM tools, and semantic search, but nothing in the evidence pack mentions [[wiki-style]] linking, backlink panels, or unlinked-mention detection — core PKM/wiki-linking features. Since Poly Mind explicitly targets note-taking/second-brain use cases, this axis is a fair question for the product, but no evidence shows it delivers wiki-links or backlinks.

   • [claimed-docs] “Poly automatically remembers key details, forming a "second brain" for you out of your files.”
   • [claimed-docs] “Bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life. Write a few thoughts, attach files, or add the broader context behind a…”

   proof ↗⚑ flag

2. power-userReference or transclude an individual block or section of one note inside another and have it stay in sync

   weight 2 · round drawn
   Anytype—none0/10

   Evidence documents only object-level linking and backlinks (docs-40, docs-28) and a live Graph of Object connections (docs-9), but nowhere describes block- or section-level transclusion/embedding that stays synced across notes — a distinct feature from whole-object linking. No mention of synced blocks, block references, or embed-with-sync exists in the pack.

   • [claimed-docs] “Linking on Anytype is what turns a bunch of separate notes into a connected knowledge base.”
   • [claimed-docs] “These connections become incredibly valuable. You can trace how ideas relate, see what's connected to a project”
   • [claimed-docs] “The Graph is a live, visual map of every Object in your Channel and how they connect to each other.”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's docs describe file search, an agent, a CLI/MCP integration, and a Markdown-based Mind/journal folder, but nothing in the evidence describes block- or section-level transclusion between notes that stays synced (no backlinks, embeds, or live-sync block references are mentioned anywhere in the docs pack).

   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life.”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life. Write a few thoughts, attach files, or add the broader context behind a…”
   • [claimed-docs] “Bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more.”

   proof ↗⚑ flag

Ownership

1. power-userMy notes live in local files or a local-first store that I own and can read without the app, and everything keeps working fully offline

   weight 3 · round to Anytype
   Anytype~partialcommunity6/10

   Anytype clearly delivers offline-first, locally-owned storage: docs state all content is stored locally, the app 'always works without an internet connection', data is encrypted at rest in a self-owned Vault, and a local-only sync mode exists that disables any external node entirely (anytype-docs-29, anytype-docs-30, anytype-docs-44, anytype-docs-31/43, anytype-docs-24). However, the underlying store is a proprietary local database/CRDT format, not plain files you can read without the app — the only path to human-readable local files is manual Markdown export/import (anytype-docs-11, anytype-docs-20), and a community report notes data 'lives unencrypted on-disk' at the index level despite vendor's encryption claims (anytype-comm-8, anytype-comm-9), showing some tension around the ownership/transparency claim. missing for 10: evidence that the live working store itself (not just exports) is human-readable outside the app, and independent confirmation that full offline use has no hidden network dependency beyond vendor docs.

   • [claimed-docs] “All your content is stored locally on your device... This means an offline-first experience: the app always works without an internet connec…”
   • [claimed-docs] “If you prefer to fully keep your data on your device without any server to sync between your desktop and mobile, you can enable local-only m…”
   • [claimed-docs] “A Vault is an encrypted container in Anytype that you own. Everything you create lives inside your Vault, stored directly on your device.”
   • [claimed-docs] “Only you can access your spaces and grant access to others — not even Anytype can see your content.”
   • [claimed-docs] “You control access. Only you can access your spaces and grant access to others — not even Anytype can see your content.”
   • [claimed-docs] “you can enable local-only mode. This mode disables Anytype's backup node entirely”
   • [claimed-docs] “Choose your preferred format: Markdown or Any-Block.”
   • [claimed-docs] “You can import content into your spaces and export it at any time.”
   • [community] “I found Anytype a few months ago and thought 'wow maybe I can ditch Turtl and just use this.' A few things bugged me: it's source-available …”
   • [community] “(Anytype team) All your data in Anytype is encrypted at rest; only the indexes are not encrypted currently. We have plans to also encrypt th…”

   proof ↗⚑ flag

   Poly!disputedcontradicted4/10

   Docs describe files/journal entries as real local files (synced Finder folder, Markdown notes in a 'Mind' folder), suggesting some local-first storage, but the core search/indexing/agent stack is cloud-hosted (remote MCP server at mcp.poly.app, OAuth-gated, server-side parsing/transcribing/indexing) — and a hands-on community reviewer explicitly notes 'your service, by nature, is reading and analyzing all of my files using a remote server,' directly contradicting a fully offline, locally-owned workflow. Missing for 10: no documentation of an offline mode, no evidence local files remain fully functional/searchable without the cloud service, and no rebuttal to the community's remote-dependency observation.

   • [claimed-docs] “your Poly files appear as a real folder in Finder or Explorer, with files downloading on demand and syncing back automatically.”
   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “It automatically parses, transcribes, and indexes your files that make it up to 20x faster for an agent to read and understand them.”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [community] “With Google Drive, I choose which files to upload. It doesn't have broad access to everything on my computer... your service, by nature, is …”
   • [community] “It feels like an LLM wrapper in search of a problem. Cloud search plus already exists... I do not find the idea of bringing my data to your …”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

2. researcherMy notes are stored or exportable in open, documented formats (Markdown, JSON) so my archive outlives the vendor

   weight 2 · round to Anytype
   Anytype✓fullprobed7/10

   Anytype explicitly documents Markdown export/import (choose Markdown or Any-Block format) and lets you import/export content at any time, plus Obsidian-style markdown migration, giving researchers a documented open-format path out of the vendor lock-in. The local API also exposes data via JSON over a documented OpenAPI spec, reinforcing programmatic, non-proprietary access to notes. missing for 10: explicit full-space JSON export/archival documentation and independent verification that exported Markdown preserves all object metadata/relations.

   • [claimed-docs] “Choose your preferred format: Markdown or Any-Block.”
   • [claimed-docs] “You can import content into your spaces and export it at any time.”
   • [claimed-docs] “Anytype will process the contents and automatically create a separate object for each Markdown file.”
   • [claimed-docs] “Open Channel Settings → Import → Obsidian and click on Import files.”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”

   proof ↗⚑ flag

   Poly~partialclaimed5/10

   Poly's Mind journal entries are explicitly stored as plain Markdown files synced to a folder (poly-docs-12), and the desktop client exposes all Poly files as a real Finder/Explorer folder that syncs back automatically (poly-docs-23), so notes aren't locked in a proprietary blob. However there is no documented bulk-export feature, no JSON export format, and no explicit statement about long-term archival or vendor-independence beyond this folder-sync mechanism; import from other note apps (poly-docs-22) is one-way and does not evidence export-out portability. Missing for 10: explicit JSON/structured export option, a documented full-archive export or backup mechanism, and any statement guaranteeing data portability/format stability if the vendor shuts down.

   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “your Poly files appear as a real folder in Finder or Explorer, with files downloading on demand and syncing back automatically.”
   • [claimed-docs] “Bring existing notes and memory archives into Poly Mind from Notion, Obsidian, Evernote, Bear, and more.”

   proof ↗⚑ flag

Self host

1. developerRun the whole system without the vendor's cloud — local-only mode or a self-hosted sync backend I operate myself

   weight 2 · round to Anytype
   Anytype✓fullcommunity8/10

   Anytype documents local-only mode (disables the backup node entirely, data stays on-device with offline-first storage) plus a self-hosted network backend option including a team-maintained Docker image for backing up/syncing without vendor cloud, and this is corroborated by the team's own community statements about p2p sync working without a central server. Some nuance remains around multi-device sync limitations in local-only mode and encryption caveats raised in community threads (indexes not encrypted, source-available license). Missing for 10: independent hands-on verification of a full self-hosted sync-server deployment (vs. docs-only description) and more detail on operational maturity of the self-host path.

   • [claimed-docs] “To back up your channels and data to a self-hosted network, you need to have a network configuration.”
   • [claimed-docs] “This mode disables Anytype's backup node entirely—meaning multi-device syncing will only work over the same local network”
   • [claimed-docs] “There's also a docker image that's maintained by the team.”
   • [claimed-docs] “you can enable local-only mode. This mode disables Anytype's backup node entirely”
   • [claimed-docs] “All your content is stored locally on your device... This means an offline-first experience: the app always works without an internet connec…”
   • [claimed-docs] “If you prefer to fully keep your data on your device without any server to sync between your desktop and mobile, you can enable local-only m…”
   • [claimed-docs] “You can even collaborate directly with others using peer-to-peer sync or self-host on your own network.”
   • [community] “(Anytype team) All your data in Anytype is encrypted at rest; only the indexes are not encrypted currently. We have plans to also encrypt th…”

   proof ↗⚑ flag

   Poly—none0/10

   All evidence points to Poly being a cloud-backed service (hosted MCP server at mcp.poly.app requiring OAuth login to Poly's cloud, cloud-based indexing/search) with no documented local-only mode or self-hosted sync backend option; community comments even complain that files/data must go through Poly's remote servers rather than staying local.

   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [community] “With Google Drive, I choose which files to upload. It doesn't have broad access to everything on my computer... your service, by nature, is …”
   • [community] “It feels like an LLM wrapper in search of a problem. Cloud search plus already exists... I do not find the idea of bringing my data to your …”

   proof ↗⚑ flag

Mobile

1. power-userFull-featured iOS and Android apps give me my whole knowledge base — capture, search, and linking — not a stripped-down companion

   weight 2 · round to Anytype
   Anytype~partialcommunity4/10

   Evidence confirms native iOS and Android apps exist and are described as polished and functional (comm-3, comm-7), and cross-device sync/offline-first architecture (anytype-docs-29, anytype-docs-30) implies capture/linking works across devices, but no documentation explicitly details mobile-specific feature parity for capture, search, and linking, and one long-term user characterizes their mobile use as 'a very basic note system,' hinting at reduced functionality vs desktop. Missing for 10: explicit docs on mobile app feature set (Queries, Graph, Types, Templates) confirming parity with desktop, and independent hands-on reviews of mobile power-user workflows beyond basic notes.

   • [community] “I've been using this for about a year. It's very polished and been stable for me. The Apple and Android apps work well too... I'm using it a…”
   • [community] “We don't offer a browser version. Instead, we have an Electron-based cross-platform app for Windows, Linux, macOS, plus native iOS and Andro…”
   • [claimed-docs] “All your content is stored locally on your device... This means an offline-first experience: the app always works without an internet connec…”
   • [claimed-docs] “If you prefer to fully keep your data on your device without any server to sync between your desktop and mobile, you can enable local-only m…”

   proof ↗⚑ flag

   Poly—none0/10

   Evidence pack extensively documents desktop, CLI, web/MCP, and Finder/Explorer integration, but contains no mention whatsoever of iOS or Android apps, mobile capture, or mobile search/linking capability. missing for 10: any evidence of iOS app, Android app, or mobile-specific capture/search/linking features.

   ⚑ flag

1. ai-native userDo everything through the API that I can do in the UI

   weight 2 · round to Anytype
   Anytype~partialprobed6/10

   Anytype ships a real local REST API (OpenAPI spec, 22 paths, confirmed by runtime probe) plus a CLI headless server and an official MCP server that let AI agents create/search/modify objects and chats — solid API-driven access. However, the API surface (22 endpoints) is far narrower than the full UI feature set (Queries/Views, Templates, Web Publishing, Graph, Types customization, import/export flows), and no evidence confirms full parity between API and UI capabilities. Missing for 10: explicit documentation or confirmation that templates, publishing, graph, and query/view creation are all API-accessible, and independent hands-on verification of parity beyond basic CRUD/chat operations.

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “Create chats, send and stream messages, react, and search chat history (anytype-heart v0.50.7)”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [claimed-docs] “The Graph is a live, visual map of every Object in your Channel and how they connect to each other.”
   • [claimed-docs] “Web Publishing lets you turn any Object into a public webpage at a URL anyone can visit.”
   • [claimed-docs] “A Query is a question you ask your knowledge base—answered instantly and kept up to date automatically.”

   proof ↗⚑ flag

   Poly~partialprobed5/10

   Poly ships a scriptable CLI (25+ commands) and a hosted MCP server exposing search, fetch, and a virtual `poly` CLI as tools, giving programmatic access to core file search/read/write/organize workflows that mirror the UI's agent capabilities (poly-docs-17, poly-probe-rt-2, poly-probe-4). However there is no documented REST/OpenAPI API (openapi probes all 404'd), and no evidence that UI-only features like sharing/permissions management, the eight folder views, flashback/version history, or journaling are exposed programmatically. Missing for 10: a general-purpose REST/OpenAPI surface, API parity for sharing/permissions, version history, and Mind/journaling features.

   • [claimed-docs] “`poly` is Poly's command-line tool: your files, searchable and scriptable from the terminal.”
   • [claimed-docs] “poly search "the contract with the March deadline"”
   • [claimed-docs] “Connect Poly to AI assistants and code editors with the local CLI first, then set up an MCP integration.”
   • [claimed-docs] “The [`poly-cli` agent skill] teaches an agent how to search, read, create, update, and organize your Poly files by composing `poly` commands…”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.poly.app/openapi.json, https://docs.poly.app/swagger.json, https://docs.poly.app/api/op…”
   • [probe] “official CLI documented at https://docs.poly.app/cli”
   • [claimed-docs] “Share a file or folder when the content belongs to you and you want specific people — or anyone with a link — to see it.”
   • [claimed-docs] “Go into flashback mode at any time to see exactly what your files looked like at any point in the past, down to the millisecond.”

   proof ↗⚑ flag

2. ai-native userExport all of my data in open formats and leave

   weight 3 · round to Anytype
   Anytype✓fullclaimed8/10

   Anytype docs explicitly support exporting content in open formats (Markdown or Any-Block) at any time, plus local-first storage, self-hosting (including a maintained Docker image), local-only mode, and exporting/importing channels into a new identity when switching network modes — giving AI-native users a documented path to export data and leave to their own infrastructure. Missing for 10: independent/hands-on verification of export completeness and fidelity, and no detailed account of exactly what metadata/relations survive export.

   • [claimed-docs] “Choose your preferred format: Markdown or Any-Block.”
   • [claimed-docs] “You can import content into your spaces and export it at any time.”
   • [claimed-docs] “export all your channels, and import them into a new identity after switching the network mode”
   • [claimed-docs] “There's also a docker image that's maintained by the team.”
   • [claimed-docs] “If you prefer to fully keep your data on your device without any server to sync between your desktop and mobile, you can enable local-only m…”
   • [claimed-docs] “All your content is stored locally on your device... This means an offline-first experience: the app always works without an internet connec…”

   proof ↗⚑ flag

   Poly~partialclaimed4/10

   Poly's desktop sync exposes files as a real folder in Finder/Explorer with files kept in their original formats and synced back automatically (poly-docs-23), and journal entries are stored as plain Markdown files (poly-docs-12), which implies some natural portability rather than lock-in. However, there is no explicit documentation of a bulk 'export all data' feature, data-export API, or guarantee that a user can fully extract their indexed memory/context in open formats and leave — missing for 10: an explicit export/data-portability feature, documentation of exporting the 'second brain'/index data, and any mention of account deletion/data takeout.

   • [claimed-docs] “your Poly files appear as a real folder in Finder or Explorer, with files downloading on demand and syncing back automatically.”
   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”
   • [claimed-docs] “Finder tags sync bidirectionally with Poly tags, colors included.”

   proof ↗⚑ flag

3. ai-native userRead the product's source under an open license

   weight 2 · round to Anytype
   Anytype!disputedcontradicted4/10

   Anytype's client code is publicly visible on GitHub (anytype-ts, anytype-heart) and even labels some AI-agent tooling 'open-source', suggesting readable source; however a hands-on community report explicitly states Anytype ships under a 'source-available license' rather than a true open license, directly contradicting an open-license claim for the core product. missing for 10: an explicit OSI-approved license file/statement for the main app repos, and resolution of the source-available vs open-source discrepancy.

   • [github] “Composable blocks: text, databases, kanban, calendar & custom Types.”
   • [github] “Extensible through a gRPC API and AI "Agents" (see AGENTS.md).”
   • [community] “I found Anytype a few months ago and thought 'wow maybe I can ditch Turtl and just use this.' A few things bugged me: it's source-available …”

   proof ↗⚑ flag

   Poly—none0/10

   No evidence of any open-source license or public source repository for Poly; it appears to be a closed, proprietary SaaS/desktop product with only documentation and API/MCP endpoints disclosed. Missing for 10: any open-source license file, public code repository, or statement of source availability.

   ⚑ flag

4. ai-native userSelf-host the core product

   weight 3 · round to Anytype
   Anytype✓fullcommunity8/10

   Anytype ships a self-hostable network node (Docker image maintained by the team) for backup/sync, plus a local-only mode that removes any dependency on Anytype's backup node entirely, and the core client itself is offline-first storing all data locally. Community input corroborates the p2p/no-central-server architecture actually working in practice. Missing for 10: independent hands-on validation of running the official self-host Docker image end-to-end, and more detail on feature parity/limits of self-hosted vs managed network.

   • [claimed-docs] “To back up your channels and data to a self-hosted network, you need to have a network configuration.”
   • [claimed-docs] “There's also a docker image that's maintained by the team.”
   • [claimed-docs] “export all your channels, and import them into a new identity after switching the network mode”
   • [claimed-docs] “This mode disables Anytype's backup node entirely—meaning multi-device syncing will only work over the same local network”
   • [claimed-docs] “you can enable local-only mode. This mode disables Anytype's backup node entirely”
   • [claimed-docs] “If you prefer to fully keep your data on your device without any server to sync between your desktop and mobile, you can enable local-only m…”
   • [claimed-docs] “All your content is stored locally on your device... This means an offline-first experience: the app always works without an internet connec…”
   • [community] “(Anytype team) All your data in Anytype is encrypted at rest; only the indexes are not encrypted currently. We have plans to also encrypt th…”

   proof ↗⚑ flag

   Poly—none0/10

   All evidence describes Poly as a hosted SaaS (cloud sync, hosted MCP endpoint at mcp.poly.app requiring OAuth, subscription-based) with no mention of a self-hosted or on-prem deployment option anywhere in the docs or probes.

   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [community] “This is really cool! I suck at organizing my filesystem and I've lost track of how many times I had to find that ONE PDF which I KNOW I have…”
   • [community] “With Google Drive, I choose which files to upload. It doesn't have broad access to everything on my computer... your service, by nature, is …”

   proof ↗⚑ flag

Metadata

1. power-userAttach typed properties and tags to notes and use them consistently for organization and retrieval

   weight 2 · round to Anytype
   Anytype✓fullcommunity7/10

   Anytype's documented object model lets users create custom Types and attach structured properties, then retrieve them via Queries that assemble results by defined criteria (docs-32/38, docs-8/21), which is the core of typed, consistent organization and retrieval; Templates further support consistent object creation (docs-13/22). Missing for 10: explicit documentation of a dedicated 'tag' property type (only 'Types'/relations are described) and independent verification that large-scale tagging stays consistent (one community report flags friction reconciling bulk/CSV data at scale, comm-10).

   • [claimed-docs] “Anytype comes with built-in Types, but you can create your own custom Types to match your needs.”
   • [claimed-docs] “you can create your own custom Types to match your needs.”
   • [claimed-docs] “A Query is a question you ask your knowledge base—answered instantly and kept up to date automatically.”
   • [claimed-docs] “Rather than manually curating a list of things you're looking for, you define the criteria and the Query assembles the result for you as a V…”
   • [claimed-docs] “A Template is a saved layout for a page that you can reuse.”
   • [claimed-docs] “you define what a good starting point looks like once and it'll form the base for every new Object you create afterwards”
   • [community] “I consistently find myself in need of a 'merge feature' when re-importing cleaned CSV data into Anytype database objects - the larger the da…”

   proof ↗⚑ flag

   Poly~partialclaimed4/10

   Poly documents tag support that syncs bidirectionally with Finder tags and colors, and search/retrieval features (natural language, visual similarity), but there is no evidence of 'typed properties' (structured metadata fields/schemas) or of consistent, structured tag-based retrieval workflows beyond basic color/tag sync. missing for 10: typed/structured property fields, evidence of consistent tag-based filtering or retrieval UI, cross-file property schemas.

   • [claimed-docs] “Finder tags sync bidirectionally with Poly tags, colors included.”
   • [claimed-docs] “Eight ways to look at a folder — Icon, Grid, Feed, Detail, Tree, C”
   • [claimed-docs] “Search inside your files with natural language that's gives you results down to the exact timestamp, page, paragraph, or frame.”

   proof ↗⚑ flag

Queries

1. power-userQuery my notes by their structured properties and render live database-style views (tables, boards, lists) from those queries

   weight 3 · round to Anytype
   Anytype~partialcommunity6/10

   Anytype supports structured Types/properties and Queries that assemble live, auto-updating Views (docs describe defining criteria and getting results as a View), plus tables/kanban/calendar-style database views per the GitHub description. However, formulas are explicitly still missing per a co-founder community comment, and there's no independent hands-on evidence confirming board/list view richness or complex cross-property querying at scale. missing for 10: independent/hands-on verification of multi-view query rendering (table/board/list) beyond docs claims, confirmation of formula/aggregation support, and evidence of performance/robustness with large structured datasets.

   • [claimed-docs] “A Query is a question you ask your knowledge base—answered instantly and kept up to date automatically.”
   • [claimed-docs] “Rather than manually curating a list of things you're looking for, you define the criteria and the Query assembles the result for you as a V…”
   • [claimed-docs] “Anytype comes with built-in Types, but you can create your own custom Types to match your needs.”
   • [github] “Composable blocks: text, databases, kanban, calendar & custom Types.”
   • [community] “Anytype has tables now, simple, no formulas yet. Plan to add formulas to databases - it's one of the top requested features on our forum (co…”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's docs describe file search, tagging, journaling, and multiple folder 'views' (Icon, Grid, Feed, Detail, Tree, etc.) but there is no evidence of structured/custom properties on notes nor of live query-driven database views like tables or boards built from property filters — the 'eight ways to look at a folder' are display modes, not queryable schema-based views.

   • [claimed-docs] “Eight ways to look at a folder — Icon, Grid, Feed, Detail, Tree, C”
   • [claimed-docs] “Finder tags sync bidirectionally with Poly tags, colors included.”
   • [claimed-docs] “Search inside your files with natural language that's gives you results down to the exact timestamp, page, paragraph, or frame.”

   proof ↗⚑ flag

Search

1. researcherInstantly search all my notes with full-text search plus filters on tags, properties, and paths — including offline

   weight 3 · round to Anytype
   Anytype~partialclaimed5/10

   Anytype's Queries let users filter by Type, tag, and other properties, and all data is stored locally for a genuine offline-first experience, which supports the filtering and offline parts of this story. However, the evidence never documents a full-text search feature, path-based filtering, or how these Query filters combine with a search index — missing for 10: explicit full-text search documentation, filtering by file/object 'paths', and any performance/scale claims for search across large note collections.

   • [claimed-docs] “A Query is a question you ask your knowledge base—answered instantly and kept up to date automatically.”
   • [claimed-docs] “Rather than manually curating a list of things you're looking for, you define the criteria and the Query assembles the result for you as a V…”
   • [claimed-docs] “All your content is stored locally on your device... This means an offline-first experience: the app always works without an internet connec…”
   • [claimed-docs] “Anytype comes with built-in Types, but you can create your own custom Types to match your needs.”
   • [claimed-docs] “you can create your own custom Types to match your needs.”

   proof ↗⚑ flag

   Poly~partialclaimed4/10

   Poly documents natural-language full-text search (down to timestamp/page/paragraph), a CLI 'poly search' command, visual similarity search, and Finder-tag syncing, but nowhere combines these into documented filter syntax on tags/properties/paths, nor claims search works offline (files instead 'download on demand' and sync back, implying online-first indexing). missing for 10: explicit support for combined tag/property/path filters in search queries, explicit documentation of offline full-text search capability.

   • [claimed-docs] “Search inside your files with natural language that's gives you results down to the exact timestamp, page, paragraph, or frame.”
   • [claimed-docs] “poly search "the contract with the March deadline"”
   • [claimed-docs] “Visual similarity — find images that look like a reference image you provide”
   • [claimed-docs] “Finder tags sync bidirectionally with Poly tags, colors included.”
   • [claimed-docs] “your Poly files appear as a real folder in Finder or Explorer, with files downloading on demand and syncing back automatically.”

   proof ↗⚑ flag

Automation hooks

1. developerDrive the app from outside — create or open notes via a URI scheme, deep links, or a local endpoint — to wire it into my scripts and workflows

   weight 2 · round to Anytype
   Anytype✓fullprobed8/10

   Anytype ships a local-API (localhost, offline) with a documented OpenAPI spec (22 paths) that developers can call directly from scripts, plus a CLI/headless server mode and an official MCP server that fronts this API for automation — all confirmed by both docs and runtime probes. This covers programmatic create/open/search of Objects for scripting and workflow integration, though it's a REST/gRPC local API rather than a classic OS-level URI scheme/deep-link handler. Missing for 10: explicit documentation of an os-level 'anytype://' URI scheme or deep-link handler, and independent third-party corroboration beyond vendor docs/probes.

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”

   proof ↗⚑ flag

   Poly~partialprobed6/10

   Poly ships a scriptable `poly` CLI (poly-docs-6, poly-docs-17, poly-probe-4) and a local/hosted MCP endpoint (poly-docs-29, poly-probe-rt-2) that let external scripts search, read, and create/edit files including Mind journal notes, satisfying the 'local endpoint for workflows' part of the story. However there is no documented URI scheme or OS-level deep-link mechanism for opening/creating notes, only CLI/MCP paths. Missing for 10: URI scheme/deep-link support, explicit CLI subcommand for creating a note by title/URI, independent confirmation of CLI scripting reliability.

   • [claimed-docs] “poly search "the contract with the March deadline"”
   • [claimed-docs] “`poly` is Poly's command-line tool: your files, searchable and scriptable from the terminal.”
   • [claimed-docs] “Connect Poly to AI assistants and code editors with the local CLI first, then set up an MCP integration.”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [probe] “official CLI documented at https://docs.poly.app/cli”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

Plugins

1. power-userBrowse and install from a large catalog of community plugins/extensions to add features the core app doesn't ship

   weight 3 · round drawn
   Anytype—none0/10

   Anytype documents an API, CLI, MCP server, and 'Agent Skill' for scripting/AI integration, but nothing in the evidence describes a community plugin/extension marketplace or catalog that users can browse and install to add features — these integrations are developer/API surfaces, not an installable plugin ecosystem.

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”
   • [github] “Extensible through a gRPC API and AI "Agents" (see AGENTS.md).”
   • [probe] “official MCP server documented at https://developers.anytype.io/docs/examples/featured/mcp/”

   proof ↗⚑ flag

   Poly—none0/10

   Evidence shows Poly ships its own integrations (CLI, MCP server, plugins for Claude/ChatGPT/VS Code) that let Poly connect into other tools, but there is no mention of a marketplace or catalog of third-party/community-built plugins that extend Poly itself with new features.

   • [claimed-docs] “Use our plugin with leading agents like Claude, ChatGPT, and VS Code and make your file workflows insanely fast.”
   • [claimed-docs] “The Model Context Protocol (MCP) lets compatible AI assistants connect to Poly.”
   • [claimed-docs] “Connect Poly to AI assistants and code editors with the local CLI first, then set up an MCP integration.”
   • [claimed-docs] “MCP clients can connect to `https://mcp.poly.app/mcp` (streamable HTTP). Authentication is OAuth: a browser window signs you into Poly on fi…”
   • [probe] “official MCP server documented at https://docs.poly.app/integrations/using-mcp”

   proof ↗⚑ flag

2. developerBuild my own plugins against a documented extension API/SDK with samples and developer docs

   weight 2 · round to Anytype
   Anytype~partialprobed6/10

   Anytype documents a local REST API (with a real OpenAPI 3.1 spec, 22 paths) plus a CLI, an official MCP server, and an 'agents skill' example showing scripts run against Anytype data — this gives developers a documented, sample-backed way to build integrations. However there's no evidence of an actual in-app plugin/extension SDK (e.g., UI extension points, a plugin marketplace, or a formal SDK package) — the API is more integration/automation-oriented than a plugin framework. Missing for 10: an official plugin SDK/framework for extending the app itself, a plugin marketplace or registry, and broader third-party plugin examples beyond MCP/agent scripts.

   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [claimed-docs] “Anytype CLI is now available - run Anytype as a headless server for automation and scripting.”
   • [claimed-docs] “enabling AI assistants to seamlessly interact with Anytype's API through natural language.”
   • [claimed-docs] “the agent can run scripts against your Anytype data — programmatically — to handle bulk edits, transformations, and repetitive tasks”
   • [claimed-docs] “Drop it into Claude Code, Cursor, Gemini CLI, GitHub Copilot, or any other agent-capable tool, and the agent can run scripts against your An…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the official Anytype MCP server (npm @anyproto/anytype-mcp 1.2.10) launched keylessly via `npx -y @anyp…”
   • [probe] “PROBE runtime (recorded 2026-09-05): the local API's machine-readable OpenAPI 3.1.0 spec (22 paths) was fetched keylessly from the vendor's …”
   • [github] “Extensible through a gRPC API and AI "Agents" (see AGENTS.md).”

   proof ↗⚑ flag

   Poly—none0/10

   Evidence shows Poly ships an MCP server, a CLI, and an agent-skill file for its own CLI, but nothing documents a first-party extension API/SDK for developers to build their own Poly plugins — the 'plugin' references are about installing Poly's plugin into other agents (Claude, ChatGPT, VS Code), not the reverse. A direct probe for an OpenAPI/Swagger spec returned 404 on all candidate paths, and no developer SDK, plugin manifest format, or extension samples are mentioned anywhere in the docs pack.

   • [claimed-docs] “Use our plugin with leading agents like Claude, ChatGPT, and VS Code and make your file workflows insanely fast.”
   • [claimed-docs] “It automatically parses, transcribes, and indexes your files that make it up to 20x faster for an agent to read and understand them. Use our…”
   • [probe] “PROBE openapi: all candidate paths 404 (https://docs.poly.app/openapi.json, https://docs.poly.app/swagger.json, https://docs.poly.app/api/op…”
   • [probe] “official CLI documented at https://docs.poly.app/cli”
   • [probe] “official MCP server documented at https://docs.poly.app/integrations/using-mcp”

   proof ↗⚑ flag

1. ai-native userChoose where my data is stored (region/residency)

   weight 2 · round to Anytype
   Anytype~partialcommunity6/10

   Anytype is local-first (data stored on-device) and offers a self-hosted network / local-only mode, giving users control over where their data physically lives rather than a fixed vendor cloud region — this indirectly satisfies data-residency concerns. However, there's no explicit multi-region cloud selection feature or documented compliance framework (e.g., EU-only hosting) — missing for 10: explicit region/residency selection UI, documented compliance certifications, and independent verification of self-host deployment satisfying legal residency requirements.

   • [claimed-docs] “To back up your channels and data to a self-hosted network, you need to have a network configuration.”
   • [claimed-docs] “There's also a docker image that's maintained by the team.”
   • [claimed-docs] “you can enable local-only mode. This mode disables Anytype's backup node entirely”
   • [claimed-docs] “If you prefer to fully keep your data on your device without any server to sync between your desktop and mobile, you can enable local-only m…”
   • [claimed-docs] “Our infrastructure handles background syncing and backups automatically, so you retain full ownership of your data without compromising your…”
   • [community] “As a teacher I just can't use Notion or any other US-service due to GDPR concerns. I hope AnyType copies as many good Notion features so I c…”

   proof ↗⚑ flag

   Poly—none0/10

   No evidence in the pack addresses data residency, region selection, or storage location choice for Poly's cloud-hosted service; documentation focuses on features, sync, and MCP/CLI integrations without any regional/residency controls. Missing for 10: any mention of data center regions, residency options, or compliance-driven storage location settings.

   ⚑ flag

2. ai-native userPrevent my data from being used to train AI models

   weight 3 · round to Anytype
   Anytype~partialcommunity6/10

   Anytype's local-first, encrypted architecture ('not even Anytype can see your content', all data stored locally, offline-first) strongly implies the vendor has no access to user data and thus cannot use it for AI training, and this is reinforced by the fact that AI integrations (MCP, local API) run entirely on localhost rather than sending data to a cloud AI service. However, there is no explicit vendor statement or ToS clause addressing AI-training use of data, and community evidence notes that index data is not yet encrypted at rest, a caveat to the 'we can't see your data' claim. Missing for 10: an explicit no-AI-training policy/ToS clause, and confirmation that even metadata/indexes are excluded from any vendor-side processing.

   • [claimed-docs] “Only you can access your spaces and grant access to others — not even Anytype can see your content.”
   • [claimed-docs] “You control access. Only you can access your spaces and grant access to others — not even Anytype can see your content.”
   • [claimed-docs] “All your content is stored locally on your device... This means an offline-first experience: the app always works without an internet connec…”
   • [claimed-docs] “Our API is now available and directly included with the desktop app, running entirely on localhost. It operates fully offline”
   • [community] “(Anytype team) All your data in Anytype is encrypted at rest; only the indexes are not encrypted currently. We have plans to also encrypt th…”

   proof ↗⚑ flag

   Poly—none0/10

   No evidence in the pack of any data-training opt-out, privacy policy, or model-training exclusion commitment from Poly; docs focus on file search/agent features, not data-usage privacy controls. Community comments raise general trust/security concerns but don't confirm or deny a training-opt-out policy, so this is an unaddressed gap rather than a dispute.

   ⚑ flag

3. ai-native userControl data retention and deletion

   weight 2 · round to Anytype
   Anytype~partialcommunity6/10

   Anytype's local-first architecture gives strong implicit control over retention: data lives encrypted at rest on-device (anytype-docs-44, anytype-comm-9), can be exported/imported at any time (anytype-docs-20), and local-only mode lets users disable all cloud/backup-node sync entirely (anytype-docs-24, anytype-docs-30), plus self-hosting for full network control (anytype-docs-6, anytype-docs-19). However there's no explicit documented 'delete my data/account' feature, data-retention policy statement, or purge/right-to-erasure mechanism in the evidence — only inferred control via local file ownership and export. Missing for 10: explicit deletion/erasure feature or policy docs, retention period controls, and confirmation of what happens to synced/self-hosted copies upon deletion request.

   • [claimed-docs] “A Vault is an encrypted container in Anytype that you own. Everything you create lives inside your Vault, stored directly on your device.”
   • [claimed-docs] “You can import content into your spaces and export it at any time.”
   • [claimed-docs] “you can enable local-only mode. This mode disables Anytype's backup node entirely”
   • [claimed-docs] “If you prefer to fully keep your data on your device without any server to sync between your desktop and mobile, you can enable local-only m…”
   • [claimed-docs] “To back up your channels and data to a self-hosted network, you need to have a network configuration.”
   • [community] “(Anytype team) All your data in Anytype is encrypted at rest; only the indexes are not encrypted currently. We have plans to also encrypt th…”
   • [claimed-docs] “Our infrastructure handles background syncing and backups automatically, so you retain full ownership of your data without compromising your…”

   proof ↗⚑ flag

   Poly—none0/10

   The evidence pack covers file import, search, sharing, versioning, and MCP/CLI integrations, but contains no documentation on data retention policies, permanent deletion, or export/purge controls for an AI-native privacy-conscious user. Community comments raise general trust/security concerns but don't confirm or deny specific retention/deletion mechanisms. missing for 10: retention policy docs, account/file deletion controls, data export or purge guarantees, any first-party statement on how long data is kept or how to fully erase it.

   • [claimed-docs] “Even conflicting edits are preserved, so you can recover any version of any file at any time.”
   • [claimed-docs] “Share a file or folder when the content belongs to you and you want specific people — or anyone with a link — to see it. You stay the owner,…”
   • [community] “With Google Drive, I choose which files to upload. It doesn't have broad access to everything on my computer... your service, by nature, is …”

   proof ↗⚑ flag

4. ai-native userOpt out of telemetry and usage tracking

   weight 2 · round drawn
   Anytype—none0/10

   The evidence pack details Anytype's local-first storage, encryption, and offline-first architecture (anytype-docs-29, anytype-docs-31, anytype-docs-43), but none of it mentions a telemetry/usage-analytics toggle or opt-out setting specifically. Privacy-by-architecture claims don't substitute for documented telemetry controls, so this applicable axis has no direct supporting evidence.

   ⚑ flag

   Poly—none0/10

   No evidence in the pack addresses telemetry, analytics, or usage-tracking opt-out settings for Poly; the docs and community discussion cover file syncing, search, MCP integrations, and general privacy/security concerns about filesystem access, not telemetry controls.

   ⚑ flag

Publish

1. researcherPublish selected notes as a browsable website — a digital garden with working links — without leaving the app

   weight 2 · round to Anytype
   Anytype✓fullclaimed8/10

   Anytype's Web Publishing feature lets users pick any Object and click Publish to generate a static HTML page on a personal subdomain (.any.org), including custom icon/cover, directly from within the app — matching the digital-garden use case with working links via the app's native linking/backlinking system. Missing for 10: no independent/community confirmation of link-following behavior between published pages or customization depth, and no evidence on bulk/multi-page publishing workflows.

   • [claimed-docs] “Web Publishing lets you turn any Object into a public webpage at a URL anyone can visit.”
   • [claimed-docs] “Pick an Object, click Publish, and Anytype generates a static HTML page hosted on your personal subdomain at .any.org/”
   • [claimed-docs] “Custom Object icon and cover — appears at the top of the page”
   • [claimed-docs] “Include a custom ANY ID — your unique handle, used for web publishing URLs”
   • [claimed-docs] “Linking on Anytype is what turns a bunch of separate notes into a connected knowledge base.”
   • [claimed-docs] “These connections become incredibly valuable. You can trace how ideas relate, see what's connected to a project”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's evidence covers sharing individual files/folders via links or team drives (poly-docs-9, poly-docs-24, poly-docs-33) and journaling/Mind notes storage, but there is no mention of publishing a collection of notes as a browsable, interlinked website (digital garden) with working cross-links. Sharing appears to be file/folder link-based, not a rendered site with navigable note-to-note links.

   • [claimed-docs] “Share a file or folder when the content belongs to you and you want specific people — or anyone with a link — to see it.”
   • [claimed-docs] “Share individual files with a link, or give a team its own storage space.”
   • [claimed-docs] “Share a file or folder when the content belongs to you and you want specific people — or anyone with a link — to see it. You stay the owner,…”
   • [claimed-docs] “Journaling gives you a quiet place to document your day-to-day life. Write a few thoughts, attach files, or add the broader context behind a…”

   proof ↗⚑ flag

Collaboration

1. researcherShare a vault or space with collaborators and edit notes together without stepping on each other's changes

   weight 2 · round to Poly
   Anytype!disputedcontradicted4/10

   Anytype's docs claim you can share a Vault/space and collaborate without sacrificing privacy, with peer-to-peer sync or self-hosting for shared editing (anytype-docs-12, anytype-docs-33, anytype-docs-39), but independent community evidence directly contradicts this: a Notion employee notes Anytype 'still lacking collaboration' and another user reports 'you can't share with others yet' as of their hands-on testing (anytype-comm-5, anytype-comm-8), and another user flags the need for a 'merge feature' when re-importing data, suggesting conflict handling for concurrent edits is immature (anytype-comm-10). Missing for 10: documented conflict-resolution/CRDT mechanics for simultaneous multi-user edits, and recent independent confirmation that sharing/collaboration works reliably today.

   • [claimed-docs] “Anytype lets you collaborate with others without sacrificing your privacy.”
   • [claimed-docs] “Collaborators don't need to provide an email address, buy a subscription, or ask anyone for permission.”
   • [claimed-docs] “You can even collaborate directly with others using peer-to-peer sync or self-host on your own network.”
   • [community] “(Notion employee) The Anytype model is really cool - they've rebuilt Lotus Notes with 21st century E2E encrypted protocols... However they a…”
   • [community] “I found Anytype a few months ago and thought 'wow maybe I can ditch Turtl and just use this.' A few things bugged me: it's source-available …”
   • [community] “I consistently find myself in need of a 'merge feature' when re-importing cleaned CSV data into Anytype database objects - the larger the da…”

   proof ↗⚑ flag

   Poly~partialclaimed5/10

   Poly supports sharing files/folders and creating team Shared Drives with members and roles, and it preserves conflicting edits so any version can be recovered, which addresses collaborative access and non-destructive conflict handling. However, this is file-versioning-based (conflicts are preserved as separate recoverable versions, not merged live), with no evidence of real-time co-editing, presence awareness, or seamless merge like a true multiplayer notes app. Missing for 10: real-time simultaneous editing support, live presence/cursor indicators, and conflict-free merge (vs. version preservation) for concurrent note edits.

   • [claimed-docs] “Share a file or folder when the content belongs to you and you want specific people — or anyone with a link — to see it.”
   • [claimed-docs] “Share a file or folder when the content belongs to you and you want specific people — or anyone with a link — to see it. You stay the owner,…”
   • [claimed-docs] “Create a Shared Drive when the content belongs to a team rather than to a person. A Shared Drive is its own storage space with its own membe…”
   • [claimed-docs] “Even conflicting edits are preserved, so you can recover any version of any file at any time.”
   • [claimed-docs] “Share individual files with a link, or give a team its own storage space.”

   proof ↗⚑ flag

Sync

1. power-userAn official sync service keeps my notes updated across all my devices with end-to-end encryption the vendor can't read

   weight 3 · round to Anytype
   Anytype!disputedcontradicted6/10

   Anytype ships an official multi-device sync via its backup-node infrastructure with claims of full E2E encryption where 'not even Anytype can see your content' (anytype-docs-31, anytype-docs-42, anytype-docs-43), and sync reportedly works well across devices (anytype-comm-9). However, a hands-on user and the vendor's own team response concretely contradict the blanket E2E claim: data is not fully encrypted on disk and 'only the indexes are not encrypted currently' (anytype-comm-8, anytype-comm-9), meaning the vendor itself confirms a gap in the encryption story it markets. missing for 10: independent security audit confirming full E2E coverage, resolution of the unencrypted-index gap, and clearer docs distinguishing at-rest vs. index encryption.

   • [claimed-docs] “Only you can access your spaces and grant access to others — not even Anytype can see your content.”
   • [claimed-docs] “Our infrastructure handles background syncing and backups automatically, so you retain full ownership of your data without compromising your…”
   • [claimed-docs] “You control access. Only you can access your spaces and grant access to others — not even Anytype can see your content.”
   • [community] “I found Anytype a few months ago and thought 'wow maybe I can ditch Turtl and just use this.' A few things bugged me: it's source-available …”
   • [community] “(Anytype team) All your data in Anytype is encrypted at rest; only the indexes are not encrypted currently. We have plans to also encrypt th…”

   proof ↗⚑ flag

   Poly—none0/10

   Poly's docs describe file sync (desktop folder sync, Finder tag sync, sharing, MCP/CLI integrations) but nowhere claim end-to-end encryption that the vendor itself cannot read; in fact the architecture explicitly routes files through Poly's servers for indexing/search, and community comments explicitly worry that Poly's remote servers read and analyze all files, contradicting the premise of vendor-blind E2EE. No evidence of an E2EE sync mechanism exists in the pack.

   • [claimed-docs] “your Poly files appear as a real folder in Finder or Explorer, with files downloading on demand and syncing back automatically.”
   • [claimed-docs] “Finder tags sync bidirectionally with Poly tags, colors included.”
   • [community] “Hooking up the Internet to my filesystem is scary. What security measures are in place to ensure a compromise of your infrastructure doesn't…”
   • [community] “With Google Drive, I choose which files to upload. It doesn't have broad access to everything on my computer... your service, by nature, is …”
   • [probe] “PROBE runtime (recorded 2026-09-14): keyless JSON-RPC initialize POST to Poly's hosted MCP server https://mcp.poly.app/mcp returned HTTP 401…”

   proof ↗⚑ flag

2. power-userView a note's version history and restore earlier versions or recover deleted notes

   weight 2 · round to Poly
   Anytype—none0/10

   No evidence in the pack mentions version history, restoring earlier versions, or recovering deleted notes; documentation covers sync, backup, encryption, import/export, and API/MCP integrations but never a version-history or undo/trash-recovery feature.

   ⚑ flag

   Poly✓fullclaimed7/10

   Poly's docs explicitly describe file-level version history and recovery: 'flashback mode' lets you see exactly what files looked like at any past point, and edits/conflicts are preserved so 'you can recover any version of any file at any time' (poly-docs-5, poly-docs-10). Since notes (journal entries, Mind files) are just files synced in Poly, this recovery/versioning applies to them too. Missing for 10: no independent/hands-on verification of the restore UX specifically for notes, and no explicit mention of undeleting a note versus just reverting content.

   • [claimed-docs] “Go into flashback mode at any time to see exactly what your files looked like at any point in the past, down to the millisecond.”
   • [claimed-docs] “Even conflicting edits are preserved, so you can recover any version of any file at any time.”
   • [claimed-docs] “New journal entries are Markdown files synced to your **Mind** folder.”

   proof ↗⚑ flag