Description
A traditional CRM is basically a database with a form, there to store customers; even the AI ones just bolt a chat box onto that form. The real work — finding out what is true and writing it down — is still left to a person. Comp AI CRM flips this: the agent is not a feature of the CRM; the CRM is where the agent keeps its notes, so the AI runs the account — researching customers, organizing information, scheduling follow-ups and executing tasks.
The reassuring part is that it never guesses to fill the database: tools report only what they observed (an email signature block, a GitHub identity), and a ledger prices the evidence — strong evidence writes to the record, weak evidence becomes a suggestion a human settles. A confidently wrong fact about a customer is worse than a blank field, because nobody can tell it is wrong. It runs as its own deployment with a work queue and schedule, so close the browser and it keeps going, resuming where it stopped after a redeploy; every page shows exactly what it did. Built on Vercel eve, MIT licensed.
Agent-run accounts: The AI agent researches customers, organizes info, schedules follow-ups and executes tasks, not just a chat helper on the side.
Evidence over guessing: Tools report only observed evidence; strong evidence writes to the record, weak evidence stays a suggestion for a human to settle, avoiding confident errors.
Background work queue: Its own deployment with a work queue and schedule, deciding what to look at next and booking its own re-checks.
Keeps going offline: Not request-response — close the browser and the agent keeps working, sessions survive a redeploy and resume from where they stopped.
Auditable actions: Every page shows exactly what the agent did, so its work is transparent and checkable.
Open and self-hostable: Built on Vercel eve durable-agent framework where tools, skills and schedules are files; MIT open source, self-hostable.
The reassuring part is that it never guesses to fill the database: tools report only what they observed (an email signature block, a GitHub identity), and a ledger prices the evidence — strong evidence writes to the record, weak evidence becomes a suggestion a human settles. A confidently wrong fact about a customer is worse than a blank field, because nobody can tell it is wrong. It runs as its own deployment with a work queue and schedule, so close the browser and it keeps going, resuming where it stopped after a redeploy; every page shows exactly what it did. Built on Vercel eve, MIT licensed.
Features
Agent-run accounts: The AI agent researches customers, organizes info, schedules follow-ups and executes tasks, not just a chat helper on the side.
Evidence over guessing: Tools report only observed evidence; strong evidence writes to the record, weak evidence stays a suggestion for a human to settle, avoiding confident errors.
Background work queue: Its own deployment with a work queue and schedule, deciding what to look at next and booking its own re-checks.
Keeps going offline: Not request-response — close the browser and the agent keeps working, sessions survive a redeploy and resume from where they stopped.
Auditable actions: Every page shows exactly what the agent did, so its work is transparent and checkable.
Open and self-hostable: Built on Vercel eve durable-agent framework where tools, skills and schedules are files; MIT open source, self-hostable.
