> ## Documentation Index
> Fetch the complete documentation index at: https://docs.3ngram.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# 3ngram documentation

> Persistent, typed memory for AI agents — decisions, commitments, blockers, and facts that survive the session.

3ngram gives Claude, ChatGPT, Cursor, and your own agents a memory that outlives the conversation. Decisions, commitments, blockers, preferences, and facts are stored as typed memories, searchable from any connected client, and never silently rewritten.

It is MCP-first: connect any MCP client to the 3ngram platform, or run the whole stack yourself with Docker Compose.

## What makes it different

<CardGroup cols={3}>
  <Card title="Append-and-supersede" icon="layers">
    Writes never destroy data. Corrections create typed edges between memories, so the old record stays queryable — including "what did I believe on date X".
  </Card>

  <Card title="Typed memories" icon="tags">
    Decisions, commitments, blockers, facts, preferences, patterns, notes, and events each keep their own lifecycle instead of flattening into chat history.
  </Card>

  <Card title="Bi-temporal facts" icon="clock">
    Facts track both when they were true in the world and when the system learned them, so retrieval can answer "what is true now" and "what was true then".
  </Card>
</CardGroup>

## The surface

The MCP server exposes 10 tools and 2 prompts over Streamable HTTP — a deliberately small surface designed around jobs to be done, not feature count. The same operations are mirrored on a REST API (`/api/v1`) for scripts, SDKs, and dashboards.

| Job                             | Tool                   |
| ------------------------------- | ---------------------- |
| Persist something worth keeping | `remember`             |
| Find what you know              | `search`               |
| Correct the record              | `revise`, `resolve`    |
| Start a session oriented        | `briefing`             |
| Carry context to another agent  | `handoff`              |
| What is currently true about X  | `get_facts`            |
| Organize your memory space      | `configure_scope`      |
| Review consolidation proposals  | `review_proposals`     |
| Inspect capabilities and config | `describe_environment` |

See the [tool reference](/reference/tools) and [prompt reference](/reference/prompts) for full schemas.

## Open source

3ngram is **open source** (Apache 2.0). The memory engine, MCP server, API, and SDK are all Apache 2.0 and fully self-hostable. The 3ngram platform runs additional proprietary cloud-operations code (billing) that is not part of this repository.

The full license map lives in [LICENSING.md](https://github.com/B3dmar/3ngram/blob/main/LICENSING.md). To get involved, see [Contributing](/contributing).

## Get started

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Connect an MCP client to the 3ngram platform and save your first memory.
  </Card>

  <Card title="Self-host" icon="server" href="/self-host">
    Run the full stack locally with Docker Compose — Postgres, Redis, and the server.
  </Card>

  <Card title="CLI reference" icon="terminal" href="/reference/cli">
    Use API-key auth from scripts and coding agents.
  </Card>

  <Card title="SDK reference" icon="braces" href="/reference/sdk">
    Call the REST-backed TypeScript client from your own code.
  </Card>

  <Card title="Memory model" icon="database" href="/concepts/memory-model">
    How append-and-supersede, typed memories, and bi-temporal facts work.
  </Card>

  <Card title="Scopes" icon="folder-tree" href="/concepts/scopes">
    Partition memories into personal, work, and custom scopes.
  </Card>
</CardGroup>