What is Engram?

Engram is a human-owned memory protocol for AI agents. It keeps durable project, team, and personal knowledge in files that humans can inspect, review, sync, and repair.

Engram is not a hidden agent brain. The agent may propose memory, but the source of truth is approved Markdown under .agents/.engram/ or an optional global memory folder.

What problem it solves

AI agents forget project decisions, repeat setup questions, and mix old context with new instructions. Built-in memory is often private to one vendor, one app, or one machine.

Engram gives memory a stable contract:

• approved facts, rules, and workflows live as Markdown
• indexes and graphs make routing fast
• writes require human approval
• hashes reveal unsafe edits
• ignore rules protect private context
• profiles isolate company, client, and personal memory so external APIs or company-provided agents do not leak context across projects
• Git gives history, portability, and team review

Mental model

Think of Engram as a knowledge memory center:

Layer	Job
Markdown	durable source of truth
JSON index	fast lookup layer
JSON graph	topic and relationship routing layer
Approval gate	trust boundary before writes
Hashes	integrity checks before reads
Ignore rules	privacy controls
Git	audit history and sync
Agent adapters	convenience, not authority

Scope priority

Engram resolves memory in this order:

1. Workspace memory: <project>/.agents/.engram/
2. Global memory: $ENGRAM_GLOBAL_DIR or engram inject --global-path <path>

Workspace memory wins. Global memory is fallback for reusable preferences and team context across projects.

Current shape

Engram includes:

• save for one approved memory
• save-session / ss for several memories from a session
• observe for raw notes that are not active memory yet
• take-control for importing existing agent guidance and docs
• graph and quality-check for review signals
• archive for wrong or superseded memory
• repair for invalid memory files skipped by index rebuild
• benchmark for retrieval regression checks
• agent skillsets, slash adapters, and MCP-style proposal tools

Next steps

• AI-agent quickstart
• Install and configure
• Human-owned protocol