Metadata-Version: 2.4
Name: exabase-langchain
Version: 0.1.0
Summary: LangChain integrations for Exabase memory
Project-URL: Homepage, https://exabase.io
Project-URL: Source, https://exabase.io/docs/api-reference
Author: Exabase
License: MIT
Requires-Python: >=3.10
Requires-Dist: exabase
Requires-Dist: langchain-core>=0.3
Requires-Dist: langchain>=0.3
Requires-Dist: langgraph>=0.3
Requires-Dist: langmem>=0.0.30
Requires-Dist: requests>=2.32
Description-Content-Type: text/markdown

# Exabase M-1 Memory Provider for LangChain

Exabase M-1 memory-provider integration for LangChain.

## About

[Exabase Memory (M-1)](https://exabase.io/memory) is a self-organising memory
engine for AI agents. It stores facts, preferences, and events, builds a living
knowledge graph, resolves contradictions, and evolves with every interaction.

M-1 is SOTA on the leading AI memory benchmark (LongMemEval), with the highest
recorded QA score, and using a small model. Read the research paper
[here](https://exabase.io/research/exabase-achieves-state-of-the-art-on-longmemeval-benchmark).


| System | Model | Score |
| --- | --- | --- |
| M-1 (Exabase) | Gemini 3 Flash | 96.4% |
| Mem0 | Gemini 3 Pro | 94.8% |
| Honcho | Gemini 3 Pro | 92.6% |
| HydraDB | Gemini 3 Pro | 90.79% |
| Supermemory | Gemini 3 Pro | 85.2% |

Exabase Memory powers memory in production apps like
[Fabric](https://fabric.so), used by 300,000+ people.

## What this integration provides

- `ExabaseToolkit`: tools for searching, adding, updating, and removing Exabase
  memories
- `ExabaseStore`: a lightweight LangGraph/LangMem-compatible store adapter
- `ExabaseRetriever`: LangChain retriever backed by Exabase memory search
- `exabase_memory_middleware`: dynamic prompt middleware that injects relevant
  memories into conversations

## Install

```bash
uv sync
```

For the package specifically:

```bash
uv run --package exabase-langchain python packages/langchain/examples/tools.py
```

## Examples

See the `examples/` folder for minimal scripts demonstrating each feature.
