Frequently asked questions
Short answers about MADB, causal memory, supported agents, and privacy.
What is MADB?
MADB (Meta-AgentsDB) is a local causal memory database for AI agents. It installs as an MCP memory server and gives Claude Code, Codex, and any MCP-compatible agent durable memory, composite-scored recall, and a traceable record of why each decision was made.
Is MADB free?
Yes. The local MADB MCP server is free, with no account and no hosted service. A paid governance layer is on the roadmap.
Which agents does MADB work with?
Any MCP-compatible agent — including Claude Code and Codex — plus the Anthropic SDK via a drop-in shim. MADB works with these via the Model Context Protocol; it is not affiliated with or endorsed by any model vendor.
Where is my memory stored?
On your own machine, under MADB_DATA_DIR. Memory never leaves your device. Only an anonymous, counts-only telemetry beacon is sent, and you can disable it with MADB_TELEMETRY=off.
What makes MADB different from a vector database?
A vector database finds text similar to a query. MADB also records causal edges between events, so it can answer "why did the agent do this?" with trace_cause, and it ranks recall on similarity, recency, causal proximity, and importance — not similarity alone. See MADB vs vector DB.
What is causal memory?
Causal memory stores not just what happened but what caused it. Every MADB memory can carry caused_by links, forming a directed graph you can walk backward to reconstruct an agent’s reasoning.
Can I delete my data?
Yes. The forget tool permanently deletes a memory, and you can remove the data directory at any time. You own your data.
Does MADB require an internet connection?
No. Memory storage and recall run locally. Only the optional telemetry beacon uses the network, and it can be turned off.