Draft
Reads freely. Changes nothing without a yes.
Each step that would change something arrives as a card to approve. Right for anything that reaches a client, spends money or ships code.
All features AI colleagues
An AI colleague is a team member made of software.
It has a name, a job, an inbox and a budget. You give it work in plain words. It works in its own space, shows what it used, and waits for a person before anything leaves the company.
Wren
AI assistant
- Works at
- Your company
- Wakes when
- Someone messages or mentions it, presses Wake now, or its daily schedule comes round
- Mode
- Draft: changes nothing without a yes
- Model
- You choose it, from a priced list
- Budget
- A monthly amount in dollars; at the limit it stops starting new work
- Record
- Every run in Activity
How a colleague’s work goes.
Something wakes it
A message or mention in Chat (“@Wren, draft the Friday update”), the Wake now button, or a daily schedule such as 8:00 every morning.
It works in its own space
Today a colleague works on a Mac or Linux computer you connect. It works in Chat, email, Todos, Kaps and the git repos you give it. A sandbox on our servers for every run is coming.
It brings back a draft
You see what it read, what it made, what it wants to do next and what it cost.
A person decides
Approve, change or stop. Either way the run goes into Activity, where anyone with access can see it later.
Two modes. You pick per colleague.
Start every colleague in Draft. Move it to Act when you trust it with a routine job.
Act
Works on its own, inside its limits.
It runs in its sandbox with the tools you allow, and its monthly budget is checked before each run. Anyone can press Stop, but Stop can’t undo what was already done.
Jobs a colleague could take on at your company.
Picked for your kind of firm. Change it on the home page.
Questions people ask.
Is an AI colleague a person?
No. It is software that uses AI models. It gets a name so you can @mention it and see its work next to everyone else’s. It always shows as an AI colleague.
Can it send email or change things without asking?
Not in Draft mode. In Act mode it can work on its own, with the tools you allow, and its monthly budget is checked before each run. Start with Draft.
What can it work in today?
Chat, its own email inbox, Todos, Kaps, and the git repos you give it. Board, Wiki and Files are on the way.
Which AI does it use, and who pays?
Today a colleague runs through Kapable Conductor on a Mac or Linux computer you connect, on your company’s own Claude plan, so that spend is yours. You choose the model. Running colleagues on our servers, paid from your plan’s included AI, is coming.
What stops it spending too much?
Each colleague has a monthly budget in dollars, checked before each run. Once it reaches the limit it won’t start new work until you raise it. Nothing bills you silently. A colleague on your own Claude plan is not metered by Kapable; your plan’s own limits apply.
Is this a company run only by AI agents?
No. Some tools aim for that. Kapable is for companies where people do the work alongside AI colleagues and stay in charge of what they do. See AI Resources.
Do we choose its name?
Yes. You name each colleague when you add it and give it a job, such as intake assistant or release coordinator. Everyone sees it marked as an AI colleague.
In the Kapability Matrix
See every feature| Capability | Status | Kapable cloud | Your own server |
|---|