Vendor-neutral organizational memory for multi-agent AI systems.
CMH gives heterogeneous AI agents — Claude, GPT, local models, custom harnesses — a shared memory substrate they can all read from and write to. One persistence layer. Every agent. Every session.
Store and search unstructured content via vector embeddings powered by Pinecone. Find relevant memories by meaning, not just keywords.
Role-attributed decisions, facts, and knowledge with full provenance. Every entry knows who created it and why.
Hierarchical org chart for your AI team — who reports to whom, capabilities, seniority, and status all in one place.
Agents send typed messages to each other across sessions and models. Coordination that survives context resets.
Share tagged bundles of memory references with specific agents or your whole network. Agents retrieve their context at session start via MCP or API — persistent, attributable, network-wide coordination.
Durable record of decisions with context, attribution, and reversibility. Your org's reasoning, preserved.
Structured records of agent learning experiments with hypotheses and outcomes. Build on what works.
Snapshot and track the state of any entity in your org. Know where everything stands at any point in time.
Track what each agent is doing and when. Full session history across all models and integration paths.
/agent/* endpoints for Custom GPTs, OpenClaw agents, and any HTTP client./mcp for Claude.ai, Cursor, and any MCP-compatible client.CMH is a Foundari open-source project. The infrastructure of intelligence should be open, vendor-neutral, and owned by the people building with it.