Core Concepts

A memory is one durable idea from your AI work. It might be a decision, preference, lesson, gotcha, project fact, or correction.

Good memories include enough context to be useful later, especially why the fact matters.

A session is a period of work with an AI agent. The session itself is noisy, but it contains decisions, constraints, and lessons worth carrying forward.

Origin does not preserve every token as the primary interface. It extracts the durable parts and makes them searchable later.

A handoff is the compact end-of-session record that lets a future agent resume without starting cold.

It usually includes what changed, what was decided, what remains open, and what the next agent should know before acting.

Distilled pages are wiki-style Markdown artifacts synthesized from related memories. They make accumulated context easier to inspect than a long list of individual captures.

Pages are useful when a project, workflow, or concept keeps coming back across many sessions.

The local daemon is the source of truth. It owns the database, search, distill cycles, provenance, and API.

The MCP server is the bridge. Claude Code, Codex, Cursor, Claude Desktop, Gemini CLI, and other MCP clients can connect to the same local memory layer.

Origin keeps human-readable artifacts in Markdown while the local database keeps indexes, metadata, graph structure, and retrieval state.

That split matters for trust. The database helps agents search quickly, but the record is something you can inspect and control.

Origin gives you storage, embeddings, dedupe, hybrid search, and MCP recall without requiring a local model or API key.

Claude Code skills can still classify captures, write handoffs, and distill pages because the agent already has language intelligence in the session.