Your AI forgets everything between sessions.
Knowns gives your AI coding assistant structured access to your project's tasks, docs, specs, templates, and memory. No repeated context. No pasted files. Your assistant works from project state, not from scratch.
npm v0.20.5
AI Workflow
You Work on task 42
AI Reading task-42: "Add user authentication"
→ Spec: @doc/specs/user-auth | Template: @template/api-auth
→ Memory: JWT pattern from task-31
✓ Generated auth middleware from template
✓ Implemented JWT validation
✓ AC verified: "Users can login with email/password"
✓ Knowledge extracted to project memory
AI Task 42 complete. 2/3 spec ACs done. Continue with task 43?
The real problem with AI coding isn't the AI. It's the context.
Without Knowns
- You paste task descriptions into every prompt
- You re-explain project conventions each session
- You copy-paste docs, schemas, and specs manually
- AI guesses at acceptance criteria
- AI forgets architectural decisions from last week
- You spend more time managing AI context than building
With Knowns
- Say "work on task 42" — AI reads the task, linked specs, and templates
- Project conventions live in memory, available every session
- Docs, schemas, and specs are structured and auto-referenced
- AI checks acceptance criteria before marking work done
- Architectural decisions persist in project and global memory
- You focus on decisions. AI handles the context.
Three steps. Then your AI reads the map.
Set up once, use every session. Your AI assistant picks up where you left off.
One command. Everything ready.
Run knowns init in your project. It creates the .knowns/ directory, generates KNOWNS.md guidelines, and configures the MCP server. Your AI assistant is connected from the start.
$ knowns init
✓ Created .knowns/ directory
✓ Generated KNOWNS.md guidelines
✓ MCP server configured
✓ AI assistant connected — ready to go
Let your AI build the context
Tell your AI assistant to create tasks, write docs, save specs, and store patterns. It uses Knowns tools directly — you don't need to learn any CLI commands.
You: Create a task for adding user authentication with JWT, link it to the auth spec
AI: ✓ Created task-42: "Add user auth"
✓ Added AC: "Users can login with email/password"
✓ Linked spec: @doc/specs/auth-spec
✓ Saved JWT pattern to project memory
Work from project state
Next session, just say "work on task 42." Your assistant reads the task, pulls linked specs and templates, checks acceptance criteria, and picks up where you left off.
You: Work on task 42
AI: Reading task-42: "Add user auth"
→ Spec: @doc/specs/auth-spec
→ Template: @template/api-auth
→ Memory: "We use JWT for auth"
✓ Implementation complete
✓ AC verified: "Users can login"
Everything your AI needs to understand your project
Structured context, not scattered files. Four pillars that turn your AI assistant from a code generator into a project-aware collaborator.
Project Memory
Your AI doesn't start from zero. Decisions, conventions, and patterns persist across sessions.
- Working memory for current session context
- Project memory for repo-level patterns and decisions
- Global memory for cross-project conventions
Structured Execution Context
Tasks with acceptance criteria, linked specs, and templates. Your AI knows what "done" looks like before writing code.
- Tasks with ACs that sync to specs
- Spec-driven development workflow
- Code generation templates linked to docs
AI Access Layer
MCP server, semantic search, and code intelligence let your AI find the right context without you pointing at files.
- MCP integration for Claude, Cursor, Kiro & more
- Semantic search with native ONNX embedding
- Code graph with symbol search and dependency tracking
Local-First Workflow
Native Go binary. Everything on your machine. Web UI with kanban, docs editor, and AI chat built on OpenCode.
- Fast native binary with built-in ONNX embedding
- Browser-based workspace with kanban, docs, and AI chat
- Import and sync docs/templates from Git repos or npm
- Workspace switching across multiple projects
And more
Import & Sync
Import docs and templates from Git repos, npm packages, or local folders. Keep them synced.
Semantic Search
AI-powered search using native ONNX vector embeddings. Find tasks and docs by meaning, not just keywords.
Code Intelligence
Search symbols, inspect code relationships, and explore a code-aware graph without leaving your local workflow.
AI Chat with OpenCode
Multi-session AI chat with real-time streaming, model switching, and optional task linking.
Workspace Switching
Switch projects in the browser while keeping kanban, docs, memory, graph, and AI chat close at hand.
Validate & Quality Checks
Check tasks, docs, and templates for broken refs, missing ACs, and SDD coverage gaps.
Why we built Knowns
I was tired of the same loop every prompt: paste docs, AI writes code, AI forgets earlier instructions, I realize I forgot to paste the new schema, start over.
I was spending more time managing the AI's memory than actually building my app.
The problem isn't the AI. The problem is the context.
So I built a tool that lets the AI read the map itself. Now, when I say "work on Task 42," I don't paste anything. The AI reads the task state, checks the docs, understands what "done" looks like. It doesn't rely on my copy-pasting skills.
When the AI solves a tricky problem, it extracts that knowledge and updates the project docs. No more lost context. No more "wait, let me give you the new file."
It just Knowns.
Built by the Community
Knowns is open source and made better by contributors like you. Join us in building the future of knowledge management.
Report Issues
Found a bug? Open an issue to help us improve.
Submit PRs
Have a fix or feature? We welcome pull requests.
Improve Docs
Help make our documentation clearer and more helpful.