# LongMem

> Privacy-first long-term memory API for AI agents and apps. Store text and files,
> search them semantically (vector + full-text hybrid), and let memory evolve over
> time (automatic fact extraction, updates/supersedes relationships, forgetting).
> Self-hostable; your data never leaves the box when run on-prem. Multi-tenant with
> strict per-tenant isolation.

Base URL: https://longmem.dev
Full reference: https://longmem.dev/llms-full.txt
Auth: every request sends `Authorization: Bearer mv_...` (your API key).
Content type: `application/json` unless uploading files.

## Core concepts

- **Collection**: a named namespace for memories (e.g. `prefs`, `people`, `docs`).
- **Memory**: a stored item (text, optionally with files/images) + metadata.
- **Search**: hybrid (vector + keyword) by default; also `vector` or `text` only.
- **Graph memory**: `/remember` extracts atomic facts and links updates/supersedes,
  so the latest truth wins; old facts can expire and be forgotten automatically.

## Quickest start (Python SDK, zero dependencies)

```python
from longmem import Longmem            # pip install longmem-sdk  (publish pending; see Source)
mem = Longmem()                        # reads LONGMEM_API_KEY
mem.add("Alex prefers dark mode", collection="prefs")
print(mem.search("what theme does Alex like?"))
```

## Key endpoints

Store a memory:
```
POST /v1/memory/collections/{collection}/memories
{ "text": "Alex prefers dark mode", "importance": 0.7, "category": "preference", "metadata": {} }
```

Search within a collection:
```
POST /v1/memory/collections/{collection}/search
{ "query": "what theme does Alex like?", "limit": 5, "mode": "hybrid" }
```

Search across all collections:
```
POST /v1/memory/search
{ "query": "...", "limit": 5 }
```

Graph memory — extract + relate facts from free text:
```
POST /v1/memory/collections/{collection}/remember
{ "content": "Alex moved from Google to Stripe last month." }
```

Build a profile summary of a collection:
```
GET /v1/memory/collections/{collection}/profile
```

List collections:
```
GET /v1/memory/collections
```

## curl example

```
curl -s https://longmem.dev/v1/memory/collections/prefs/search \
  -H "Authorization: Bearer $LONGMEM_API_KEY" -H "Content-Type: application/json" \
  -d '{"query":"what theme does the user like?","limit":5}'
```

## MCP (Model Context Protocol)

LongMem speaks MCP at `POST /v1/mcp` (JSON-RPC) with tools `search_memory`,
`get_memory`, `store_memory`, `list_collections` — point an MCP-capable agent at it
with your bearer key.

## One-prompt agent onboarding

Paste this into Claude Code or Cursor to wire memory into your project:

> Use the LongMem memory API at https://longmem.dev. Read the bearer key from the
> LONGMEM_API_KEY env var. Before answering, search relevant memory with
> `POST /v1/memory/search {"query": <user intent>}`. After learning a durable fact
> about the user or project, store it with
> `POST /v1/memory/collections/project/memories {"text": <fact>}`. Prefer the
> `longmem` Python SDK if available.

## Links

- Docs: https://longmem.dev/docs.html
- Source: https://github.com/11data/longmem