We just open-sourced SupaVector: a self-hosted memory and retrieval platform for AI applications and agents.
SupaVector gives AI systems a persistent memory layer so they can:
- ingest documents and notes
- store and retrieve relevant context
- run grounded Q&A over stored knowledge
- manage longer-term memory in a more structured way
The current stack combines:
- a C++ vector store for similarity search
- a Node.js gateway for APIs, auth, docs UI, and jobs
- Postgres for metadata, auth, tenant settings, and memory state
A lot of AI apps need more than one-shot prompts. They need a reliable way to keep context over time, retrieve the right information when needed, and stay deployable inside your own environment.
SupaVector is our take on that: a self-hosted memory engine you can run yourself and plug into your agents, backends, or apps.
Right now, the public repo is focused on:
- single-node self-hosted deployment
- your own environment
- your own model/provider credentials
It also includes:
- a CLI for setup and local operations
- support for write / search / ask workflows
- provider support centered on OpenAI, with Gemini and Anthropic support where available
GitHub: https://github.com/Emmanuel-Bamidele/supavector
We’d really love feedback—especially on the developer experience, architecture, and where this could be most useful in real AI products.