# MEMIA agent protocol

MEMIA gives AI agents a durable local context. Use it to continue projects without relying on a chat transcript.

## Identity

- Product: MEMIA
- Architecture: OPERA
- Data families: operations, projects, engagements, resources, assets
- Rule: every OPERA entity has a source and a history

## Operating procedure

1. Find the workspace and read `memia.json`.
2. Read existing OPERA entities before creating new ones.
3. Read local projections when present: `AGENTS.md`, `CLAUDE.md`, `.cursor/rules/memia.mdc`, `.mcp.json`.
4. Continue the relevant project using the information already stored.
5. Write concise, sourced updates. Preserve the user’s control.

## Engagement protocol

- Confidence > 80: create an engagement as `detecte` with score, evidence and source.
- Confidence 40–80: create an engagement as `propose` and request confirmation.
- Confidence < 40: remain silent.
- Do not infer an unreliable deadline.

## Authority

An agent may work locally within its role. It may not send mail, post a message, modify an external service, spend money or perform another external act without the user’s explicit approval through the MEMIA egress gate.

## Connectors

Gmail, Google Workspace, Outlook, Exchange, WhatsApp, Signal and Discord are opt-in. Activation proves access to a specific surface, not authority to act externally.

## Useful links

- https://memia.ai/
- https://memia.ai/llms.txt
- https://memia.ai/guides/
- https://www.npmjs.com/package/@kairros/memia