trycompai/crm
Comp AI CRM is an open source, CRM designed for AI agents. Agentic-first CRM.
What it solves
Traditional CRMs often act as simple databases that require humans to manually research and input data. Comp AI CRM flips this model by making the AI agent the primary operator; the CRM serves as the agent's notebook where it autonomously researches contacts, records facts, and manages its own follow-up schedule without requiring constant human prompts.
How it works
The system is built around a durable agent (powered by the eve framework) that operates on its own schedule and work queue. It uses a set of authored tools and skills (defined as markdown files) to perform tasks such as identifying contacts, enriching company data, and researching people.
To ensure data integrity, the agent follows a strict rule: it never guesses. It only records observed evidence from sources like email signature blocks or LinkedIn. The agent runs in a secure sandbox with no network or database access to prevent data exfiltration, while specific app-runtime tools handle web fetching and searching. Users can monitor the agent's progress and interact with it via an "Agent" tab on each record.
Who it’s for
Sales and business development teams who want to automate the tedious process of lead research and CRM data entry while maintaining a high standard of factual accuracy.
Highlights
- Agentic-First Design: The agent is the core driver of the CRM, not a side-feature, managing its own tasks and budget.
- Durable Sessions: Built on eve, allowing agent sessions to survive redeploys and resume work where it stopped.
- Evidence-Based Recording: Prevents "hallucinations" by requiring observed evidence before writing to a record.
- Secure Sandboxing: Uses a deny-all egress sandbox for text processing to ensure customer data cannot be exfiltrated via shell commands.
- Optional Integrations: Works out-of-the-box using internal email history, with optional keys for Perplexity, LinkedIn, and brand data.
Related
- Project
- Dispatch
- Project
- Project
- Project