crm: Run an Agent-First CRM for AI Agents

Overview

Comp AI CRM is a TypeScript CRM built around eve, Vercel's filesystem-first framework for durable agents. It addresses the limits of CRMs that leave research and record-keeping to people: its agent works through a queue on its own schedule, researches contacts and companies, and records findings as evidence rather than guessed facts.

It is a self-hosted, single-tenant web application for teams that want an agent to maintain CRM records while keeping people involved when evidence is uncertain. The agent can run without external research API keys, and the project includes a web app, API, and separately deployed agent.

Key Features

• Durable agent sessions and scheduled work, backed by a database queue with expiring leases.
• Tools for searching CRM history, identifying contacts, researching people, enriching companies, recording facts, and scheduling follow-ups.
• Evidence-based record updates: strong evidence can update records, while weaker findings become suggestions for human review.
• An Agent tab on contact, company, and deal records for conversations, progress, and questions requiring a person's input.
• Optional service connections with read-only or read-and-write access, reviewed tools, and per-agent grants.
• Gmail and calendar integrations, with mailbox sync; Microsoft mail access is read-only and forward-only.
• A sandbox for agent work with network egress denied and no database credentials passed to the shell.
• A TypeScript monorepo using Next.js, NestJS, Prisma, and PostgreSQL, with Bun and Turborepo.

Use Cases

• A sales team wants an agent to research contacts and keep records current, while escalating uncertain identity matches for review.
• A small team wants follow-ups scheduled from CRM work queues instead of relying on a person to run recurring research manually.
• An organization wants to connect mailboxes for CRM context but restrict Microsoft access to reading messages and importing only newly checked mail.
• A self-hosting team wants a single-tenant CRM it can deploy and configure itself, including its identity provider and agent integrations.

Getting Started

The local quick start uses Bun and Docker Compose:

git clone https://github.com/trycompai/crm.git
cd crm
cp .env.example .env
bun install
docker compose up -d
bun run db:deploy
bun run dev

Set the required authentication and sign-in values in .env first. See the README for OAuth setup, configuration, and deployment details.

Alternatives

• twenty: Twenty is a configurable CRM for technical teams, while Comp AI centers on scheduled AI research and evidence-based record updates.
• crm: Frappe CRM focuses on conventional lead, deal, and interaction workflows with calling and messaging integrations, rather than autonomous record research.
• atomic-crm: Atomic CRM organizes contacts, follow-ups, notes, and deals with React and Supabase, without Comp AI's scheduled agent-driven enrichment.

Considerations

• The project is young, so its recent activity and releases are encouraging but do not establish long-term operational maturity. It has handled at least 18 issue closures in the measured period, with a median close time of 40 days.
• Local development requires Bun, Node.js 22 or newer, Docker Compose, and PostgreSQL. Production deployment involves three separate services plus a database.
• Automated tests are not indicated in the supplied project requirements, so teams should assess their own verification needs before adopting it.
• External research and integrations depend on configuration. The agent can work without optional research keys, but that limits its sources.
• The project is single-tenant by design, which may not suit organizations that need separate customer workspaces.