Repository navigation
Initial public foundation for RealBrain, a local-first operating-memory toolkit for AI agents.
Highlights
- Typed models for events, neurons, synapses, beliefs, dream runs, and workspace items.
- SQLite operational store.
- Markdown/Obsidian adapter with path confinement.
- Optional GBrain adapter with markdown fallback.
- Host-agent tool layer in
realbrain_server.tools. - Global workspace activation.
- Dream engine with hypothesis/suggestion-only outputs.
- Extraction, synapse hygiene, contradiction review, curiosity queue, and nightly consolidation loops.
- Fake-data demo and unit tests.
- Public docs for architecture, security/privacy, roadmap, launch, contribution, and outreach.
Safety boundaries
RealBrain is not consciousness, not autonomous authority, and not a hosted memory service. Host runtimes own identity, approvals, secrets, and external actions. Dream/consolidation output must not be treated as fact without evidence review.
Looking for contributors
- MCP server wrapper.
- OpenClaw plugin package/example.
- Adapter examples.
- Benchmarks.
- Demo GIFs/screenshots.
- PyPI packaging.
- Docs and safety tests.