Relationship memory, on your machine
A practical guide to the experimental Agent CRM CLI for shell-capable agents.
Agent CRM is a local-first, schema-aware relationship database for shell-capable AI agents. It is a CLI backed by SQLite, not a hosted app or an agent itself. There is no account, CRM server, synchronization service, or runtime network dependency.
A short path to your first record
Section titled “A short path to your first record”- Install the npm package and try a temporary database.
- Select an agent profile and bind its database.
- Search, remember, and prepare with context.
- Read the privacy boundary and establish tested backups before using real data.
What you get
Section titled “What you get”People, organizations, interactions, and follow-ups are built in. The schema is not limited to those defaults: create your own objects, add typed fields, and connect records with directed relationships to model the CRM your work needs. FTS5 search, immutable mutation history, and exact-retry idempotency apply across the same local store.
This makes Agent CRM adaptable across sectors and workflows without pretending every organization uses the same pipeline. The supported building blocks remain explicit: eight field types, optional formats and validated relationships rather than arbitrary executable schema logic.
The CLI reference, CSV guide, backup guide, and privacy guide are rendered from the repository’s authoritative documents at build time—not separately maintained copies.
What you do not get
Section titled “What you do not get”No hosted dashboard, MCP server, automatic duplicate merge, embeddings, or cloud sync. Your agent host may still send retrieved data to a model provider or messaging channel. Local storage is not a promise that all agent activity is private.