Independent, and precise about our words

Not affiliated with Meta Platforms

Meta-Agents.AI LLC is an independent company. We are not affiliated with, endorsed by, sponsored by, or connected to Meta Platforms, Inc. or any of its products or services — Facebook, Instagram, WhatsApp, Threads, or the Llama models. Any resemblance in name is coincidental to those products.

The “Meta” in Meta-Agents refers to meta-agents — higher-order agents that create and refine other agents. It’s a term of art about agents building agents, not a reference to the social-media company. Our work is memory infrastructure for autonomous agents, and it is our own.

What we build

We make MADB (Meta-AgentsDB): local, durable, causal memory for AI agents. It runs on your machine, remembers across sessions, records why things happened, and lets an agent replay the exact chain behind a decision. It works with Claude, Codex, and any MCP-compatible agent, and the core is free in one command.

The vocabulary we coined

We introduced a specific language for agent memory. We use these terms precisely and consistently, and this is what we mean by them:

Causal memory / the causal field. Memory that records not just what was stored but what caused what — a graph of memories linked by their causes, rather than a flat pile ranked by similarity. The “causal field” is that graph seen whole.

caused_by. The edge that links a memory to the earlier memories that led to it. It is captured in the moment a decision is made, not inferred afterward.

Decision lineage. The complete causal chain behind an action — the record that lets someone explain and defend why an autonomous agent did what it did.

trace_cause. The operation that walks that chain: ask “why did this happen?” and get the exact lineage back, replayed rather than reconstructed.

The Cortex. The live 3D map of a store’s causal field — writes enter, flow through the shared spine, and surface as recall, every memory traceable back to its cause.

These terms are ours. We’ll keep defining them clearly here so the words stay precise as the category grows.

Where we’re headed

The trajectory we’re building for is one where agents increasingly design, tune, and deploy other agents. That future inherits one hard requirement at every layer: it has to remember what it did and be able to say why. MADB is built to be that substrate — the causal memory and decision-lineage layer underneath. Enforcing policy on top of that record is on the roadmap; the causal record itself is live today. Read the longer argument in The 2030 Memory Problem.