Dakera
Self-hosted memory engine for AI agents.
Long-term memory that survives sessions, in one Rust binary on your own servers.
Dakera gives AI agents a memory. Store what your agents learn, recall it by meaning and by keyword, and let it fade or strengthen with use: hybrid vector and full-text retrieval, cross-encoder reranking, entity extraction, sessions and importance decay, behind a REST, gRPC and MCP API. Everything runs on your own infrastructure, CPU-only if you like, with no external AI service.
docker run -p 3000:3000 -e DAKERA_ROOT_API_KEY=dk-mykey ghcr.io/dakera-ai/dakera:latestModels
Dakera runs its models locally with ONNX Runtime. This organisation holds verified copies of every model Dakera downloads (byte-for-byte identical to the originals, credited to their authors under their own licences, checked against the same SHA-256 whichever source a server downloads from) and Dakera's own int8 builds of the GLiNER entity extraction models.
- ๐ง Embedding: bge-large-en-v1.5 (default), bge-small-en-v1.5, e5-small-v2, all-MiniLM-L6-v2, bge-m3, modernbert-embed-base, gte-modernbert-base
- ๐ฏ Late interaction: answerai-colbert-small-v1
- ๐ Reranking: bge-reranker-v2-m3 (default), bge-reranker-base
- ๐ท๏ธ Entity extraction (int8, built by Dakera): gliner-bi-small (default, English), gliner-x-small (default, multilingual), gliner-bi-edge, gliner-medium, gliner-multi
- ๐๏ธ Intent detection: paraphrase-multilingual-MiniLM-L12-v2 (forget requests, standing instructions)
- ๐๏ธ Speech to text: whisper-tiny.en, whisper-base.en, whisper-tiny, whisper-base (default), whisper-small
- ๐ผ๏ธ Image and page retrieval: colmodernvbert
Build with Dakera
- ๐ Self-hosting (Docker, Kubernetes, Helm): dakera-deploy, dakera-helm
- ๐ MCP server for Claude, Cursor and other agents: dakera-mcp
- ๐งฐ SDKs: Python, TypeScript, Go, Rust
- ๐ฆ Frameworks: LangChain, LangChain.js, LlamaIndex, CrewAI, AutoGen, Vercel AI SDK, Strands
- โจ๏ธ Command line: dakera-cli
Get in touch
Questions, feedback or a feature you need: amine@dakera.ai