Introducing crmkit

We're introducing crmkit, an open-source CRM built for AI agents to drive directly. It is agent-first and headless by design: there is no dashboard to click through. The agent loads a one-page operating manual and runs the CRM itself over a plain HTTP API. Responses are plain, grepable text by default - one labeled line per record, JSON on request - so they stay token-cheap and survive context truncation.

crmkit ships as a single static Go binary. With pure-Go SQLite embedded, a CGO_ENABLED=0 build produces one file you drop on a server and run; point it at Postgres instead when a deployment grows. There is no UI to host, no SDK to install, and no schema to wire up before an agent can start remembering people, companies, deals and support tickets.

Underneath the simple surface is a real system of record. Contacts, companies, deals, tickets and activities are all workspace-scoped for multi-tenancy. Every record gets a short, stable handle the agent threads through later calls, a free-form custom object so the schema bends without server changes, tags for grouping, and full-text search across everything in one call. Writes are protected by optimistic concurrency, and an append-only audit log doubles as record history - every change is captured as a field-level diff you can scope to a single record.

It is built for more than one agent. Several agents - and the teammates alongside them - work the same records at the same time, so crmkit gives them consistent reads, safe concurrent writes, and a clear trail of which agent or member did what. Due and overdue follow-ups surface through a reminders endpoint the agent pulls at the start of a session, so nothing falls through the cracks. Recent additions include first-class support tickets with their own conversation timeline, so the same backbone runs a sales pipeline or a support desk.

crmkit speaks the Model Context Protocol, so ChatGPT and Claude - and any ChatBotKit agent - connect to it as a one-click connector and get the CRM as a clean set of tools. First login is passwordless: an emailed code mints a bearer token, and the agent is off. For a worked example, the AI Market Bot uses crmkit as the shared system of record for the companies and competitors it tracks over time.

crmkit is open source under the MIT license and lives at github.com/crmkit/crmkit, with the project home at crmkit.ai. Download the binary, run one migration, and hand your agent its first workspace.