Feature request: Hindsight as memory implementation

[nicoloboschi]

It would be a great improvement to have Hindsight as memory provider since it's state of the art and fully local.

Oh My Pi uses it a memory backend for example
https://hindsight.vectorize.io/blog/2026/06/08/oh-my-pi-hindsight-memory

I can help with the integration

[sosidudku1]

Thanks for the proposal and for offering to help, and congrats on the LongMemEval results. To set expectations honestly: Atomic ships its own local 5-layer memory and that stays the default. The memory layer currently does not expose a provider interface, and building that abstraction is not on our near-term roadmap. If you want to move this forward anyway, the most useful artifact would be a short design sketch of the integration surface you would need from us: hooks, and what data flows in and out. We would read it with interest, but no commitments on timeline from our side. One hard requirement for any eventual integration: a fully documented local-only path (ollama or llama.cpp plus local embeddings), since local-first is the core promise of this project.