Axonode

Graph-structured long-term memory layer for Retrieval-Augmented LLMs
Work-in-progress — API not yet stable

Why Axonode?

• Neural–style graph storage – typed nodes & edges mimic axons/dendrites, not flat key–value blobs.
• Token-budget optimisation – automatic chunking, clustering and summarisation keep context windows small.
• Pluggable RAG pipeline – slot in OpenAI, Anthropic, Llama 2/3, etc., behind a single retrieval interface.
• Embeddings-agnostic – use any vector backend (Faiss, PGVector, Pinecone, etc.).
• Python-first – minimal deps, zero-config pip install axonode to get started.

Quick start

pip install axonode==0.0.0a0   # placeholder version

python -m axonode.cli ingest ./docs/*.pdf
python -m axonode.cli chat

⚠️ Alpha quality – the commands and import paths will change.

Roadmap

How it works (30 seconds)

┌── Ingest ──┐   ┌──────── Graph store ───────┐   ┌─ Retrieval + Context ┐
│  Files     │ → │  Nodes: Chunk | Concept   │ → │  Prompt w/ citations │
│  Chat logs │   │  Edges: TAGS | RELATES_TO │   └──────────────────────┘
└────────────┘   └───────────────────────────┘

Contributing

1. Clone the repo and poetry install / pip -e .
2. Run make test (pytest + coverage).
3. Open a PR against main.

See CONTRIBUTING.md for coding-style & commit-message guidelines.

License

This project is licensed under the MIT License – see LICENSE.

Citation

If Axonode helped your research, please cite:

@software{axonode_2025,
  author       = {Erez Azaria},
  title        = {Axonode},
  year         = 2025,
  url          = {https://github.com/AXONODE-DEV/axonode},
  version      = {0.0.0a0}
}

Questions? Open an issue or join our Discord: discord.gg/XXXXXX