Know and Own your data, context and intent.
Portable Markdown vault, search by meaning, and agent-ready MCP and CLI.
Your data stays yours.
Precise faster fetch, fewer tokens—fetch only what agents need. Streamlined requests, lower cost.
Self-host from the repo, or use our hosted setup—ready when you are.
Knowledge scatters across tools and chats. Models need context—but dumping everything in a window wastes tokens and hurts accuracy. The bet here is notation you own plus retrieval that returns the right slice: precise faster fetch, fewer tokens, lower cost.
What matters is rarely one document—it is how decisions, rationale, and history connect. When that map lives only in people’s heads or inside a vendor’s closed session, you lose continuity when teams change or when you switch tools.
Long context alone is not enough. Agents need high-signal segments: scoped by project, tag, time, and cause. Over-fetch wastes money; precise retrieval—--limit, --fields, then get-note only for what matters—saves tokens and keeps answers sharp.
A vault of Markdown you control, indexed for search, callable from any agent via CLI or MCP. Tiered retrieval: narrow scope, cheap first (paths or count), fetch full content only when needed. Import from the tools you use; keep authority in files you can copy, diff, and move.
CLI: Connect via CLI or MCP
knowtation search, get-note, list-notes. MCP: same tools in Cursor/Claude. AGENT-INTEGRATION.md (in repo)Retrieval & token levers
--limit, --fields path, --since/--until, --count-only. Tiered retrieval = fewer tokens. RETRIEVAL-AND-CLI-REFERENCE.md (in repo)
Agents often lack intention and an overarching view over time. Knowtation helps.
--since, --until to scope results to date ranges.
Track decisions and events across time. Optional links: follows, caused by.
Chunk, note, episode, project — reason at the right granularity.
All of these use optional frontmatter: date, project, tags, causal_chain_id, entity, episode_id, follows. Set them in the Hub (New note full form) or via CLI/API. Not auto-extracted—you or your agent add them. Filters: --since, --until, --chain, --entity, --episode. Full list and behavior: INTENTION-AND-TEMPORAL.md (in repo) and How to use in the Hub.
No servers to run. Sign in with Google or GitHub; your vault lives on the Internet Computer. Back up to GitHub anytime.
Markdown + frontmatter (Obsidian-style). Single source of truth. Use any editor.
Web UI to view and search the vault, manage proposals, and review imports. Self-host or use hosted when available.
Search, get-note, list-notes, index, write, export, import. Same interface for you and agents.
ChatGPT, Claude, Mem0, NotebookLM, Google Drive, MIF, audio/video transcripts.
One vault, many projects. Filter by project or tag. All content available when you need it.
Audio and video → one note per recording. Smart glasses and wearables supported.
Cursor, Claude, DeerFlow, Windsurf, MCP-Agent — any runtime that speaks CLI or MCP.
Capture and import into the vault → index → query from agents with tight fields and filters → refine notes → index again. Each cycle improves what retrieval can see; nothing depends on a single model release or a single vendor’s memory.