AI & agents
An assistant that has read the whole project.
Not a chatbot bolted onto the sidebar with no idea what you're working on. It has your documents, threads, to-dos and files as context — and, when you ask it to, it can act on them.
Three jobs
Answer, draft, do.
Most AI features stop at the first one. The useful part is the third — and the reason it works is that it's inside the tool where the work actually lives.
Answers, with receipts
Ask anything about a project and get an answer that links back to the document, message or to-do it came from. If it can't find the answer in your project, it says so rather than inventing one.
Drafts in your voice
Client updates, replies, release notes, proposals, proofreading. It reads the existing writing in the project and matches it, so what comes back sounds like your team rather than a press release.
Agents that do the work
"Turn this thread into a to-do list." "Schedule the review for the day before launch." "Write the handover doc." It proposes the exact changes; you approve them before anything is written.
How it works
Context in, proposal out, you decide.
It reads the project
Documents, threads, comments, to-dos, calendar and the files you've pointed it at — scoped to the one project you're in.
You ask in plain English
No prompt engineering, no slash commands to memorise. Ask the way you'd ask a colleague who'd been on the project all year.
It shows its working
Answers cite their sources. Actions are shown as a specific list of changes before anything happens.
You approve, or you don't
Nothing is created, edited or sent without a click from you. Every AI action is attributed and reversible.
Your model, your key
Claude or GPT. Per project.
Different teams trust different models, and the good one changes every few months. So it's a setting, not a lock-in. Pick the model per project, switch whenever you like.
- Anthropic Claude and OpenAI GPT models, selectable per project
- Bring your own API key on the Unlimited plan and pay the provider directly
- Turn the AI off entirely, per project or account-wide
- Your content is never used to train anyone's models
What it never does
- Send anything to a model provider you haven't asked it to work on
- Act on a project without an explicit approval
- Read a project you're not a member of
- Post as you without saying it was drafted by AI, unless you edit and send it yourself
In practice
Things people will actually type
About the AI
The questions we get asked most about how this part works.
Which AI model does it use?
What exactly is an AI credit?
Do you train AI models on my projects?
Can the AI change my stuff without asking?
Something we haven't answered? Ask us directly — a person replies.
Try it on your own projects.
Join the waitlist. Tell us what you'd want it to do and we'll build towards it.
Free to join. One email when your invite is ready — nothing else, ever.
