AIGX Foundation

What it is

AIGX Brain is an external memory genome stored as plain-text .aigx files. It extends the AIGX method beyond codebases: one addressed index, one bounded compiler, one auditable memory system that any agent can use.

Not "store more". Store better. The win is addressed retrieval and bounded context packs, not vector top-k dumps.

Why it wins

Addressed retrieval. Task scope resolves through brain.index.aigx, then ranks inside that scope.
Typed memory policies. Facts, preferences, rules, episodes, and procedures have different write/confirm/forget behavior.
Portable ownership. The user can export, inspect, diff, and move the brain.
Auditability. Every read, write, and forget is attributable to a connection.
Safety by default. Sensitive memory never auto-writes and never auto-injects.

Default brain structure (all files are `.aigx`)

brain/
├── soul.aigx
├── brain.index.aigx
├── profile.aigx
├── preferences.aigx
├── people.aigx
├── work.aigx
├── personal.aigx
├── episodes.aigx
├── procedures.aigx
├── rules.aigx
└── manifest.aigx

soul.aigx is always loaded. brain.index.aigx is the keystone that maps task scopes to memory ids. Everything else is shardable by domain and principal.

MCP-first integration (the real plug-and-play)

Connectors are the product surface. A beginner should connect their agent with one MCP endpoint and one connect code in under 5 minutes.

Tool: aigx_recall(task, budget_tokens?)
Returns: { context_pack, ids, gaps }

Tool: aigx_remember(observation)
Returns: { status: written|proposed, id?, reason? }

The server decides auto-write vs approval based on memory type, sensitivity, confidence, and conflict. Agents stay simple and safe by default.

Minimal API surface

POST /v1/brains - create a brain and dashboard link.
POST /v1/brains/:id/connections - mint a connect code with scopes.
POST /v1/recall - compile bounded context pack for a task.
POST /v1/remember - propose or write memory using policy gates.
POST /v1/forget - propose memory removal.
GET /v1/export.aigx - export full brain genome.

Agent protocol

Recall before personal answers.
Respect precedence: rule > fact > preference > inferred.
If gaps report unknowns, ask the user.
Remember sparingly - durable and useful observations only.
Never overwrite on conflict - propose and queue for approval.

Launch standard for aigx.dev

Public route /brain plus raw canonical /aigx-brain.md.
MCP quickstart with copy-paste setup and connect code.
Scoped auth model: default is context.read + memory.propose.
Export and audit available from day one.
A first-time user connects and gets visible recall in under 5 minutes.

The full design proposal and implementation details live in docs/brain-protocol.md on GitHub ↗.

A plug-and-play brain for any agent.