Cognitive infrastructure for AI agents.

We build the substrate AI agents run on — memory, reasoning scaffolds, verifiable state. Starting with the hardest part: memory that doesn't collapse under context-window limits, and that an agent can actually reason over.

Engrava — the memory database for AI agents.

An embedded cognitive graph database for AI agents. Graph memory (7 edge types), hybrid FTS5 + vector search, algorithmic dreaming (4-signal consolidation, 3 gates), tamper-evident audit trail. No Neo4j. No Redis. Just SQLite.

pip install engrava

MIT licensed · engrava.ai · docs

• Code-first. Every claim is backed by implementation. No whitepapers without running code.
• Research posture. Grounded in consciousness research, not AI hype. Citations where it matters.
• Open by default. MIT licensed, public roadmap, discussions over DMs.

• Studio: sovantica.ai
• Engrava: engrava.ai
• Contact: hello@sovantica.ai
• Security: security@sovantica.ai — see SECURITY.md