Donna GitHub →

Motivation

How Donna started.

A real problem inside our own agency. A tool we built to solve it. A vision that kept expanding once we saw what it could do.

1. The pain we lived with.

We run an agency. Like most agencies, our biggest job is to build the right thing for the client, not the thing the client said in the kickoff, but the thing they actually need three weeks later when reality has changed twice.

That meant our engineering team needed context. Real context. The kind that lives in a Fathom recording from last Tuesday, an email reply nobody copied to the team channel, a Google Doc someone edited at 11pm, a WhatsApp message from the client's PM, a Linear ticket comment from yesterday.

The context was always there. It was just scattered. And every time engineering had to make a real decision, someone had to spend half a day reassembling it. Multiply that across five clients and twenty engineers, and you have a quiet, expensive friction that nobody puts in the budget.

We weren't missing information. We were missing a place where the information could live together.

2. The first solve, for engineering.

We built a tool that pulled everything into one organised vault. Meetings, mail, drive files, chats, tickets, all normalised into a single shape, organised by who, what, when, and how things connected. Our engineers used Claude Code, which could now read from that vault and write code that reflected what the business actually decided.

It worked. The friction dropped. Code shipped that actually matched what the client meant. We had finally bridged the gap between business and tech, the one every agency complains about and almost nobody fixes.

3. The second solve, for the business side.

Then a question kept nagging at us: why is this only for engineering?

The same context layer that helps engineers ship the right code can help sales write the right proposal, help ops write the right runbook, help an executive assistant draft the right follow-up email. The friction isn't unique to tech teams, it's everywhere there's a gap between "what the business decided" and "what someone is about to write or say."

So we built it so the engineering team could create governed documentation , under strict rules, with provenance, and the business team could read it, refine it, and use it as a real source of truth. Not a wiki nobody updates. Not a Notion graveyard. A living layer with shape.

We built an MCP server so any coding agent on any engineer's machine could plug into Donna and generate or update that documentation, under whatever rules the team wants to enforce. The substrate is the same. The audience is wider.

4. The third solve, for everyone else.

Then the bigger question: most companies in the world aren't tech companies. They sell services. They consult. They run a clinic, a studio, a small business with five employees and twenty years of know-how locked in someone's head.

What if those companies could have AI colleagues, designed agents that actually know the company, the offer, the customers, the way decisions get made? Not a chatbot that hallucinates. A teammate that draws on years of real material the company has been quietly accumulating, and uses it to support marketing, sales, finance, operations.

What if a solopreneur could run a company where most of the "team" is a handful of well-designed agents that hand work to each other like real collaborators , taking decisions, analysing the market, helping the human stay strategic instead of buried in execution?

That's the version of Donna we want to ship. The chat workspace where the colleagues happen to be partly human and partly AI, and you can't always tell from the answer which is which, because both are working from the same well-kept knowledge.

5. Why open. Why self-hostable.

Your company's knowledge is the most valuable thing you own. It's also the thing you should never rent. We made Donna open source so you can read the code, and we made it self-hostable so you can run it on infrastructure you control. Cloud and on-premise run the same code path, different deployment, identical product.

Bring your own OAuth credentials. Bring your own storage bucket. Bring your own language model. The substrate stays inside your perimeter unless you choose otherwise.

6. Where we are.

Donna is early. The foundations are shipped, multi-tenant chat, the connector framework, the Bronze ingestion layer for Fathom and Gmail. The next pieces are the ones that turn it from "a chat tool with integrations" into "a chat workspace with agent colleagues":

The full sequence is on the roadmap. None of it is shipped yet, we're building in the open, and we'd rather show you the honest state than over-promise.

The product is a chat workspace. The bet is that the chat is just the surface , and the real value is the layer underneath that keeps your company's knowledge working for you.