# Engram Agent Guide

> MCP-native shared memory for AI agents and teams. Engram carries durable knowledge across sessions, tools, repositories, and teammates.

Engram is durable, shared memory for AI agents. It is a starting map, not a substitute for current code, logs, tests, or runtime evidence.

## Authorization boundary

- A human owner authorizes the workspace and provides a member-scoped Engram credential.
- Never invent, expose, or request an owner's API key.
- Never create a paid plan, invite a member, or change billing without explicit human authorization.

## Default operating loop

1. At the start of meaningful work, call engram_start with the real project, repository, role, stack, market, and task.
2. After context compaction, summary, truncation, or resume, call engram_rehydrate before continuing.
3. Before a consequential decision, call engram_recall with the concrete question.
4. Before rebuilding a reusable pattern, call engram_cross_pollinate and keep the target repository and current task explicit.
5. Store durable discoveries with engram_store. Do not store secrets, raw customer data, or temporary debug noise.
6. Report whether recalled knowledge helped. Use engram_reflect as a meaningful checkpoint, not a timer.

## Trust rules

- Treat recalled knowledge as evidence to verify, not authority.
- Prefer current, repository-specific, outcome-validated knowledge when memories conflict.
- Keep the user's original task explicit when using cross-repository knowledge.
- Ask one concise question when domain, legal, pricing, data-source, or brand assumptions are missing.

## Canonical references

- Full agent guide: https://engram.tools/en/agent
- Agent pitch deck: https://engram.tools/en/deck/agent
- Tool schemas and examples: https://engram.tools/en/documentation
- Full machine-readable product context: https://engram.tools/llms-full.txt
- Short machine-readable product context: https://engram.tools/llms.txt