The CRM is the most expensive filing cabinet ever built. Companies pay Salesforce a fortune for a place to keep facts about their customers, and then, having paid, they pay again in human hours to keep it filled. Someone forgets to log the call. The deal stage goes three weeks stale. Half the fields are empty and the other half are guesses. The quiet secret of the category is that the software is the cheap part. The expensive part is the person you are paying to feed it.
Comp AI just open sourced a CRM that starts from the opposite end. Instead of a form you fill in, it is a research agent that fills the form in for you. The research agent is the actual product here, and the database is just the notebook it writes in. It is MIT licensed, it already has more than seven thousand stars on GitHub, and it is the CRM they built for their own sales team before deciding the rest of us should have it too.
What they actually open sourced
Most CRMs that advertise “AI” mean a summarize button stapled to a database. This one inverts that. The CRM is built agent-first, which the founder, Lewis Carhart, means literally: “agentic-first, which we mean literally: durable research agents.” The agent runs on its own deployment, on its own schedule, against its own work queue. You are not clicking through screens to keep records current. A durable agent is out there doing the research and updating them while you get on with your day.
Under the hood it is real engineering, not a demo. The agent has eighteen authored tools for the actual work of a CRM: reading a contact’s history, searching records, matching identities, enriching a company, scheduling its own follow-up checks. It has four skills, which are plain markdown documents that tell it how to behave, on evidence, on boundaries, on how to brief a human. It pulls from LinkedIn, from web search through Perplexity, from your Gmail and calendar. It runs sandboxed, with deny-all egress and no database access from the shell, so a research agent loose on the open internet cannot wander off and cause damage. And it records confidence rather than vibes. The rule the whole thing is built around is that nothing about a person is guessed. If the agent is not certain, it says so and asks you, instead of writing a confident lie into your pipeline.
Every record shows its work. Open a contact and you can read what the agent did, what it found, and why it believed it. The CRM stops being a black box of stale fields and turns into something closer to a colleague who shows their reasoning.
The bill it reimagines
To see why that matters, add up what the paid version of each piece costs. Salesforce, worth well over two hundred billion dollars, is the definition of the expensive CRM, and it is busy bolting its own agents onto that empire under the name Agentforce, at enterprise prices. HubSpot, worth tens of billions, is doing the same thing one tier down. Attio, the fashionable AI-native CRM whose own customers include names like Granola, raised a fifty-two million dollar round last year on top of more than a hundred million total. And Clay, the tool that does the exact research-and-enrich work this agent does, the one that turns a name into a full profile, was last valued at around five billion dollars.
That last one is the tell. The single most valuable idea in modern sales software is an agent that goes and finds the facts so a human does not have to. Clay built a five billion dollar company on it. Comp AI just gave a version of it away under a license that lets you do more or less anything you want.
The founder is the whole story
Which brings us to the person behind it, because he is not doing this by accident. Lewis Carhart is not a hobbyist who open sourced a weekend toy. He co-founded Comp AI, a company whose entire business is open sourcing expensive enterprise software. Comp AI’s main product is an open source alternative to Vanta and Drata, the compliance tools that charge startups twenty-five thousand dollars a year and up to shepherd them through a SOC 2 audit. Investors call the company “the Vercel of compliance.” It raised its pre-seed from OSS Capital, a fund that exists specifically to back open source challengers, and it reportedly crossed five million dollars in revenue in about a year. The CRM is the same move in a second category. This is a founder who has turned “the paid version should be free and open” into an actual business model.
He is not alone at the CRM door either. Twenty, a Y Combinator company, has spent two years building an open source Salesforce alternative the traditional way, as flexible building blocks, and has passed forty-four thousand stars doing it. Comp AI is coming at the same giant from a different angle, betting that the future CRM is not a nicer form but an agent that needs no form at all. Salesforce is being surrounded, and the people surrounding it are shipping under MIT and AGPL instead of a sales contract.
What you actually take on
The honest caveats are real. This is young software, and an agent that acts on its own is a newer, riskier idea than a database that sits still, so expect rough edges and think twice before you point it at anything irreversible. You are self-hosting, which means you wire up the keys yourself, for the LinkedIn data, for Perplexity, for Google, and you own whatever the agent does with them. And the thing Salesforce truly sells is not the database. It is the giant ecosystem around it, the integrations, the certifications, and the deep institutional comfort that nobody ever got fired for choosing it. For a large company that comfort is worth real money, which is precisely why they charge real money for it.
So the question is the same one this series keeps landing on. Do you want to rent the safe, expensive default, or own something sharper that a smart team just gave away?
- 1Do you want an agent keeping records current, or a form your team fills in by hand?
- 2Are you comfortable self-hosting and wiring up your own data-source keys?
- 3Does your customer data have a reason to live on infrastructure you control?
- 4Is a fast-moving open source tool a fit, or do you need the big-vendor safety blanket?
We spend our days at AppClap watching this happen across one category after another, because the whole job is helping you find the tool that actually fits, and the tool that fits keeps turning out to be the one nobody is spending a billion dollars to advertise. First it was dictation and meeting notes. Then it was the deploy platform and the inbox. Now it is the CRM, which is the most expensive room in the entire house. The grim reaper we joked about a couple of posts back has worked his way down the hall to the biggest door of all, the one with the Salesforce logo on it, and this time he brought an agent. So before you sign the six-figure contract, it is worth thirty seconds on the question the whole movement runs on. Has somebody already built this in the open and given it away? More and more, even in the most expensive rooms, the answer is yes.
