v2.1.0-rc2 · code frozen, QA audit in progress

The auditable memory + cognition substrate for AI agents

Grounded recall you can prove. Simulation you can fence. mem20 separates real memory from imagined memory at the storage layer — and enforces that separation with a contamination firewall.

97
MCP Tools
13
Tool Domains
0.0
Contamination Rate
MIT
License

Memory that earns trust through proof

Today's AI agents hallucinate, confabulate, and forget — not because they're broken, but because their memory systems treat every stored token as equally real. mem20 fixes that by partitioning memory at the engine level:

🧠

Grounded Memory

Facts derived from real observations, user statements, or events. Indexed, searchable, and always attributable.

💭

Simulated Memory

Hypotheticals, counterfactuals, and imagined scenarios — fenced off from recall indexes, never contaminating grounded truth.

🛡️

Evidence-Based Promotion

Simulated records become grounded only when paired with real-world prediction resolution — not free-text self-assertion.

🔍

Audit Engine

Run memory_audit_contamination() anytime. A clean system reports rate 0.0 — and the firewall makes that invariant structural.

MCP-Native

Speaks Model Context Protocol over stdio. Drop into Hermes, Claude, Cursor, or any MCP host. 97 tools, 13 domains.

🔌

Pluggable Integrations

Blender and Unity ship as optional modules. Add Figma, Unreal, or anything else via the blank integration template.

Four layers, one hard invariant

mem20 is not a wrapper around a vector DB. It's a purpose-built memory engine with protocol, cognition, storage, and governance layers — all enforcing a single rule: simulated content never reaches grounded indexes.

The Invariant

Vector and BM25 indexes are grounded-only by construction. _add_to_vector_index() and _add_to_bm25_index() call _assert_grounded() on the lowest write path. Contamination is impossible regardless of which tool invoked the write.

LayerWhat it doesWhere it lives
TransportMCP server, JSON-RPC 2.0 over stdiomcp/mcp_server.py
CognitionPlan, chain, reason, reflect, working memorycog/cognitive_engine.py
MemoryGrounded vs simulated store, vector + BM25 indexesmemory_engine/memory.py
GovernanceContamination guards, epistemic veto, auditEnforced at the store engine level

Built for teams who can't afford confabulation

🤖 Multi-Agent Fleets

126 bots, isolated namespaces, cross-contamination rate 0%. Persistent per-bot memory that survives restarts and model swaps.

🏗️ Long-Running Agents

Working memory that persists across sessions. An agent that remembers what it was doing yesterday, last week, last month.

🧪 Research & Simulation

Run counterfactuals without polluting grounded recall. Promote simulations to facts only when evidence supports it.

Ready to give your agents real memory?

Join the beta. Get early access, direct input on the roadmap, and a channel to the team building it.

Sign Up for Beta →