Blame for README.md

---
license: mit
base_model: BAAI/bge-m3
pipeline_tag: feature-extraction
library_name: onnx
tags:
- dakera
- ai-agent-memory
- agent-memory
- long-term-memory
- rag
- self-hosted
- mcp
- onnx
- mirror
---

<p align="center">
  <a href="https://dakera.ai"><img src="https://dakera.ai/assets/logo-128.png" width="96" alt="Dakera logo"></a>
</p>

<h1 align="center">bge-m3</h1>

<p align="center"><b>Multilingual embedding model for Dakera · a verified mirror by Dakera, the self-hosted memory engine for AI agents</b></p>

<p align="center">
  <a href="https://dakera.ai"><img alt="Dakera: dakera.ai" src="https://img.shields.io/badge/Dakera-dakera.ai-c9a227?style=flat-square"></a> <a href="https://dakera.ai/docs"><img alt="Docs: dakera.ai/docs" src="https://img.shields.io/badge/Docs-dakera.ai%2Fdocs-0b0b0b?style=flat-square"></a> <a href="https://github.com/Dakera-AI"><img alt="GitHub: Dakera-AI" src="https://img.shields.io/badge/GitHub-Dakera--AI-181717?style=flat-square&logo=github&logoColor=white"></a> <a href="https://huggingface.co/DakeraAI/bge-m3/blob/main/LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/License-MIT-2b6cb0?style=flat-square"></a> <a href="https://huggingface.co/BAAI/bge-m3"><img alt="Original: BAAI/bge-m3" src="https://img.shields.io/badge/Original-BAAI%2Fbge--m3-6b7280?style=flat-square"></a>
</p>


> An unmodified, checksum-verified copy of the files [Dakera](https://dakera.ai) uses from [BAAI/bge-m3](https://huggingface.co/BAAI/bge-m3). The model is by **BAAI**; all credit goes to its authors.

## In Dakera

Embeddings for about 100 languages and long documents: `DAKERA_MODEL=bge-m3`.

Dakera fetches this embedding model from this repository first and from the original repositories if it is unreachable, so installs never depend on a single source. Both are checked against the same SHA-256, and the [Dakera Docker image](https://github.com/Dakera-AI/dakera-deploy) ships the default models ready to run, with no download at startup.

## Files

Copied byte for byte from:

| File | Source | Size |
|---|---|---|
| `tokenizer.json` | [BAAI/bge-m3](https://huggingface.co/BAAI/bge-m3) | 17.1 MB |
| `onnx/model_quantized.onnx` | [Xenova/bge-m3](https://huggingface.co/Xenova/bge-m3) | 569.7 MB |

<details>
<summary>SHA-256 checksums</summary>

```text
21106b6d7dab2952c1d496fb21d5dc9db75c28ed361a05f5020bbba27810dd08  tokenizer.json
0826f8c1ab9edf1801db86c61919d4d108e8bfc0b809ec823ad366882ff0b77d  onnx/model_quantized.onnx
```

</details>

## Licence and credit

MIT, the same as the original model (see [`LICENSE`](https://huggingface.co/DakeraAI/bge-m3/blob/main/LICENSE)). For what the model does, how it was trained and how to cite it, see the [original model card](https://huggingface.co/BAAI/bge-m3).

## About Dakera

**[Dakera](https://dakera.ai) is a self-hosted memory engine for AI agents.** It gives agents long-term memory that survives sessions: hybrid vector and full-text retrieval, reranking, entity extraction and importance decay, in a single Rust binary that runs on your own servers, with an MCP server for Claude, Cursor and other agents.

```bash
docker run -p 3000:3000 -e DAKERA_ROOT_API_KEY=dk-mykey ghcr.io/dakera-ai/dakera:latest
```

| | |
|---|---|
| 🌐 Website | [dakera.ai](https://dakera.ai) |
| 📚 Documentation | [dakera.ai/docs](https://dakera.ai/docs) |
| 💻 GitHub | [github.com/Dakera-AI](https://github.com/Dakera-AI) |
| 🚀 Self-hosting (Docker, Kubernetes, Helm) | [dakera-deploy](https://github.com/Dakera-AI/dakera-deploy) |
| 🔌 MCP server | [dakera-mcp](https://github.com/Dakera-AI/dakera-mcp) |
| 🧰 SDKs | [Python](https://github.com/Dakera-AI/dakera-py) · [TypeScript](https://github.com/Dakera-AI/dakera-js) · [Go](https://github.com/Dakera-AI/dakera-go) · [Rust](https://github.com/Dakera-AI/dakera-rs) |
| 💼 LinkedIn | [Dakera AI](https://www.linkedin.com/company/dakera-ai) |

Questions or feedback: [amine@dakera.ai](mailto:amine@dakera.ai).