Meta‑AgentsDB / the cortex — how meta‑agents think

Your agents' memory, as one causal field.

This is the cortex. Every step your agents take in Claude Code or Codex lands here — the decision, the reasoning, and what it caused. Writes enter on the left, flow through the shared MADB spine, and surface on the right as recall — so agents act on remembered, auditable context instead of guessing, and every action stays traceable back to its cause. Drag to turn it.

drag to rotate · hover a node

ENTRY · writes · roots in → the causal spine → OUTBOUND · recall · effects out

Sample This is example data. Point MADB at your own work and this becomes your cortex — every Claude Code & Codex decision, the reasoning behind it, and what it caused, on one auditable spine. Fast-track building with clear traceability. claude mcp add madb -- uvx madb-mcp-server

the lobes

One substrate, every surface you run

Each lobe is a real cluster of tenant scopes in your engine. They don't touch each other — they converge on the shared causal spine.

what it holds

The shapes your memory thinks in

Event types seen across the store — the recurring primitives the engine reasons over.

how it fires

Recent activity

Operation mix from the most recent window — write & recall dominate; causal walks trace the “why”.

the substrate

The physical engine underneath

A local-first LSM store: writes land in a WAL, flush to memtables, compact into leveled SSTables, indexed by a causal DAG and a vector cache for semantic recall — all on this disk, no cloud.