Most CRMs right now are trying to add AI to their product. That's not the same thing as being an AI-native CRM. HubSpot, GoHighLevel and the rest built their software long before AI went mainstream, so AI ends up bolted onto the side of a system that was designed for clicking.
I'm on a mission to fix that. I'm building an AI-first CRM where you set up and run the whole system by chat, and the UI is the backup instead of the main event. I recorded a demo of where it stands today, and this post walks through what I showed and why I'm building it.
Three problems with today's CRMs
I've spent years setting up CRMs for clients, mostly GoHighLevel and HubSpot. These are the three things I keep running into.
- Configuration is all point-and-click. To change anything, you have to know how their system works, where every setting lives, and then go change it by hand. You should be able to configure the whole system by telling it what you want.
- They're closed source. I want to be able to extend and modify the source however I see fit. In 2026, software that runs your business really should be open source.
- You don't truly own your data. Yes, there are APIs. But you rarely get the full picture of everything going on in your account. As AI becomes part of how every business runs, not being able to get at your own data, today or ten years from now, becomes a real problem.
What is a headless CRM?
A headless CRM is one where chat is the primary way you build and change the system, and the UI is secondary. That's the code name I'm using for now. It's probably a terrible name, and honestly it may not even end up being just a CRM.
It's built on two simple ideas.
- Types. If you know databases, types are basically tables. Three primitive types ship with every account: contacts, pipelines and opportunities. I'm keeping those as small as possible because I don't want to make assumptions about how your business works. Anything else you need, you create, just like custom objects in HubSpot or GoHighLevel.
- Events. Every create, update and delete on any type is an event. Your workflows subscribe to those events and fire when they happen. Even actions like creating a workflow get logged as events, so you always have the full history of what happened in your system.
The UI also doesn't look like a typical SaaS dashboard with a sidebar on the left. I want it to feel more like an operating system, with a taskbar across the bottom for contacts, pipelines, workflows, events and types.
The demo: building a CRM without clicking a single setting
For the demo I spun up a new profile (think GoHighLevel sub-account) for a made-up company, Johns Construction. Everything below was done by talking to the assistant. I was dictating with Wispr Flow, so most of the time I wasn't even typing.
Contacts in one sentence
I asked it to create 10 example contacts with first names, last names and emails. They showed up in Contacts a few seconds later, and each one was logged as a contact.created event.
A pipeline it knew to ask about
Next I asked it to add Jordan Lee to our leads pipeline. There wasn't one yet. Instead of failing, the assistant noticed that and asked if I wanted to create it. I told it to make a leads pipeline with four stages: lead, qualified lead, proposal and sale. It built the pipeline and dropped Jordan in.
Workflows by description
Then I described two automations.
- Proposal alert. When a deal moves to the proposal stage, send me an internal email. Then I said "I just had a good call with Jordan Lee, move him to proposal," and the email landed in my inbox.
- Post-sale drip. When a deal moves to sale, send a three-email sequence: a thank-you, an intro to me as their account manager, and a review request. The assistant asked what timing I wanted before building it. I said one minute between emails for the demo.
The workflows show up as a visual flow you can inspect. You can still drag a deal between stages by hand if you want to. You just don't have to.
Each account gets "tools," starting with email and text messages. Email currently runs on Resend under the hood, but the provider is meant to be swappable for AWS, GCP or whatever you prefer. SMS isn't hooked up yet.
Dashboards on the fly
After moving a few more contacts into the pipeline, I asked for a pie chart of opportunities by stage, and it appeared on the dashboard. Bar charts, line charts and time series work the same way. They're basic for now, but it's real reporting you get just by asking.
Custom types
Finally, I asked for a new type called Purchase. Each purchase belongs to a contact, a contact can have many purchases, and each one has an amount in cents and a description. It built the type with the relationship and required fields. Then I had it create some example purchases and asked for a chart of revenue per customer in descending order.
There was one small bug: the chart wasn't converting cents to dollars. That's not the end of the world, but it's a good reminder that this is early.
The bigger idea: write a spec for your business and hand it off
The demo builds things one prompt at a time, but that's not the end goal. Where I want this to go is this: you write one spec describing how your business works, what data you collect and how deals move, and hand it to the assistant. It then creates every type, field, pipeline and workflow you need, maybe in a few phases rather than one giant prompt.
UI is a second-class citizen here. It's still important for viewing and interacting with your data, but it's not how you build or change the system.
On top of that core, there will be tools and plugins, from me and eventually other developers, maybe in some kind of marketplace. You don't buy a bloated CRM with a thousand features you'll never touch. You get the core infrastructure and add what your business actually needs.
Open source, and you own your data
Two things are non-negotiable for me.
- It will be open source. You can read it, extend it and change it however you see fit.
- You own your data. Host it with me, or spin it up on your own servers and your own database. Either way, every event and every record is yours, now and in ten years.
It's not usable by the public yet. Everything is running locally while I build it out, but open sourcing it is the plan.
FAQ
What is an AI-native CRM?
An AI-native CRM is designed around AI from day one, so you set up and run it by talking to it. Most CRMs today add AI features onto software that was built for point-and-click configuration.
What does "headless CRM" mean?
It means the UI isn't the main way you control the system. Chat is how you build and change things, and the interface is there to view and interact with your data when you need it.
Is this an open source alternative to HubSpot or GoHighLevel?
That's the direction. It covers the CRM core (contacts, pipelines, opportunities, workflows, reporting), and it will be open source and self-hostable. It isn't a feature-for-feature replacement today.
Can I use it now?
Not yet. It's running locally while I build it. If you want early access, see below.
Want early access?
I'm looking for a handful of business owners and agencies to try it early and tell me what's missing. If you want in, request access and I'll reach out when there's something you can get your hands on.
I'm really excited about this one. I think there's a real need for a CRM that's open source, lets you own your data, and is truly AI-first. You shouldn't need to be a HubSpot or GoHighLevel expert to run your business. You should be able to just say what you want done.