workthin is a knowledge management service for AI-assisted development. When someone on your team solves a problem, workthin records the solution. Next time anyone encounters the same problem, their AI tool finds it automatically.

A vaccine protects you from diseases others already fought. workthin does the same for software problems — one person solves it, everyone gets the immunity. That's why we call it Knowledge Vaccination.

Engineers, tech leads, POs, PMs — anyone who uses AI tools for work. Engineers record technical solutions, POs record decision rationale, tech leads record architecture choices. All of it becomes searchable by the team's AI.

ChatGPT Memory is personal and locked to one tool. workthin works across Claude Code, Cursor, VS Code, Codex CLI, Gemini CLI, and more MCP-compatible tools, can be shared with your team, and persists even when people leave the project.

Mem0 provides memory for AI applications but requires developer integration and starts at $249/month. workthin works out of the box with any MCP-compatible tool, structures knowledge as problem-solution pairs, and starts free.

Wikis require someone to write and someone to search. workthin records knowledge automatically from AI conversations and surfaces it automatically when a similar problem appears. No manual writing, no manual searching.

One command: npx workthin@latest init. It handles authentication, project detection, MCP configuration, skills/commands, AGENTS.md, and optionally installs the CLI globally. Already have wt installed? Run wt init directly. See the Quick Start guide for details.

Claude Code, Claude Desktop, ChatGPT, Cursor, VS Code / GitHub Copilot, OpenAI Codex CLI, GitHub Copilot CLI, Gemini CLI — 8 dedicated integrations. Any other MCP-compatible tool can also connect manually. See MCP Integrations for setup instructions per tool.

MCP (Model Context Protocol) is a standard that lets AI tools connect to external services. workthin runs as an MCP server, so your AI tool can search and create knowledge directly within your conversation. No copy-pasting needed.

It authenticates you, detects the git remote to auto-create/join a project, asks which AI tool you use, configures MCP for that tool, generates AGENTS.md with workthin instructions, and creates /workthin and /workthin-search commands if your tool supports them.

/workthin captures the current session's solution as knowledge. /workthin-search finds existing knowledge before you start solving a problem. Both are generated by wt init for supported tools. See Skills & Commands for the full list.

Yes. Each person needs to run wt init once to set up their own MCP connection (authentication is per-user). But the project is auto-detected from git remote — no invitation needed. AGENTS.md and skill files are committed to git, so those are shared automatically.

Yes. If they use an AI tool that supports MCP, they run wt init just like engineers. They can record decisions, specifications, and reasoning as knowledge. Their AI will also search existing knowledge when relevant.

When you record knowledge, AI automatically extracts a structured format: problem, environment, attempted solutions, and working solution. Tags are auto-generated. You don't need to format anything manually.

workthin uses vector similarity search (OpenAI text-embedding-3-small) with hybrid reranking. Results are scored by semantic similarity, environment fingerprint match, tag overlap, and freshness. Your environment's results get a boost.

workthin records the environment where knowledge was created — runtime, framework, dependencies hash, OS. When you search, results from similar environments are ranked higher. This helps when a solution works in Node 20 but not Node 22.

You can link related knowledge entries using prev_knowledge_id. For example: "Auth design" → "Token management" → "Refresh implementation". This preserves the flow of reasoning across multiple sessions.

Branches let you fork a knowledge entry using branch_from_id. If a solution has a variant for a different environment (e.g., "Linux version" vs "macOS version"), you branch instead of overwriting.

Yes. You can add comments (supplements, warnings, deprecation notices), update the body or scope, and change tags. Comments are visible to everyone who can see the knowledge. Deprecation comments mark the entry as outdated.

When you create knowledge without specifying a scope, workthin saves it as personal and returns a suggested_scope with an AI-recommended scope and reason. Your AI tool will ask if you want to change it — for example, promoting personal knowledge to global if it's universally useful.

personal (only you can see it), project (visible to project members), and global (visible to all workthin users). Personal and project reads are always free. Global reads are limited on the Free plan.

Projects are auto-detected from your git remote. When you run wt init in a repository, workthin creates or joins a project based on the remote URL. Everyone who runs wt init in the same repository joins the same project automatically.

Not for engineers — running wt init in the same repository auto-joins the project. For non-engineers without git access (POs, designers), an admin can invite them by email from the project settings. The invited person needs a workthin account.

Their knowledge stays in the project. This is one of workthin's core benefits — knowledge doesn't disappear when people move on. The next person's AI can still find and use everything that was recorded.

Yes. If you're not in a git repository or don't have a project, everything defaults to personal scope. You can still use global scope to share knowledge with the community.

Free ($0), Pro ($10/month or $100/year), and Enterprise (custom pricing). All plans include the full CLI, MCP server with all supported AI tools, AI structuring, and 3-stage secret protection. See the Pricing page for a full comparison.

Unlimited knowledge storage, 30 global reads per month, full CLI and MCP access, AI-powered structuring and tagging, 3-stage secret protection, and semantic search. The only limit is global reads.

A global read is counted each time your AI searches or retrieves a global-scoped knowledge entry. Personal and project-scoped reads are always free and unlimited, regardless of your plan.

Everything in Free, plus unlimited global reads, custom security rules per project, and priority support. $10/month or $100/year (save 2 months).

Everything in Pro, plus SSO/SAML, SLA guarantee, and a dedicated account manager. Contact hello@workthin.com for pricing.

No trial period. The Free plan is the trial — use it as long as you want with 30 global reads per month. Upgrade to Pro only when you need more.

No. Only human users need accounts. Your AI agents use workthin through MCP or CLI without additional cost, regardless of how many tools you connect.

Contributing to the global library earns additional reads. Publishing +1.0, being viewed +0.1/view, referenced in search +0.3, resolving someone's problem +1.0, being branched +0.5. Rewards are instant and accumulate monthly. See the Concepts docs for the full table.

Yes. Downgrade to Free and your data remains. You'll be limited to 30 global reads/month but personal and project knowledge stays accessible.

No. workthin uses the OpenAI API for structuring and embedding. OpenAI's API policy states: “OpenAI does not train on data sent via the API by default.” See the OpenAI data policy and our Security page for details.

A 3-stage pipeline runs before any knowledge is stored. Stage 1: 20+ regex patterns detect known secret formats (AWS, GitHub, Stripe, JWT, etc.). Stage 2: Shannon entropy catches unknown high-randomness strings. Stage 3: AI classification handles ambiguous content. Custom rules can be added per project on the Pro plan. See Secret Protection docs for details.

Supabase (PostgreSQL) in the US. All data encrypted at rest with AES-256 and in transit with TLS 1.3. See the Security page for the full list of sub-processors.

No. Personal scope is enforced at the database level with Row Level Security. Only you can access your personal knowledge. Project and global scopes are shared as configured.

workthin uses OAuth 2.1 for MCP HTTP connections — the same standard used by Stripe, Sentry, and Supabase MCP servers. Claude Code handles the OAuth flow automatically. The CLI also supports API key authentication for programmatic access. See MCP Setup docs for details.

Yes. Account deletion removes all your data permanently — knowledge, comments, resolves, API keys, and profile. You can request deletion from account settings.

SSO/SAML is available on the Enterprise plan. It's currently in development. Contact hello@workthin.com to discuss your requirements.

wt init (setup), wt create (record), wt search (semantic search), wt get (retrieve by ID), wt update, wt list, wt check (duplicate check), wt resolve (mark as helpful), wt comment, wt login/logout, wt config. See the CLI Commands reference for full details.

Yes. All operations available through CLI and MCP are also available via REST API at /api/v1/. Authentication via Bearer token (API key). See the API Endpoints reference for the full list.

Locally in ~/.workthin/config.json. The server only stores a SHA-256 hash of the key — the plaintext key is never stored server-side.

Yes. Set the WORKTHIN_API_KEY environment variable in your CI pipeline and use wt search or wt create from scripts. The CLI works non-interactively when environment variables are set.

Your API key was invalidated (e.g., after a database reset or key regeneration). wt init automatically re-authenticates — follow the browser login prompt. No manual action needed.

Run /mcp in Claude Code and select "Authenticate" or "Reconnect". workthin uses OAuth 2.1, so the first connection requires browser authentication. After that, tokens are cached.

Possible reasons: the search threshold may be too high for your query (try more specific terms), the knowledge may not have an embedding yet (wait a moment after creating), or the knowledge is in a different scope. Search with scope: personal or scope: all to check.

Auto-tagging uses the OpenAI API. If the request fails (rate limit, network issue), tags default to empty. The knowledge itself is still saved. You can add tags manually via wt update.

MCP sessions are currently in-memory. Restarting the dev server clears all sessions. Run /mcp → Reconnect in Claude Code to re-establish the connection. This is a known limitation being addressed with Redis-backed sessions.