It works where your work already is
Connect the documents, tools, and accounts a job needs. Each connection records who may use it and what it may do.
AI coworkers your team can trust with real work.
They learn how your company works, use whichever AI is best that week, and never send, pay, or change anything without asking.
macOS 13 or later. Intel Mac build.
Trusted by teams atNorthwindArdentKestrelHalcyonMeridian
They learn your company.What they learn stays yours.
Download for macOSWhat it is
One place for your context, your agents, and your team.
Everything a job needs, reachable by whichever model you picked this month, with a real boundary between what is yours and what is shared.
Show it once, keep it forever
Walk an agent through a task and it saves the shape of it as a skill your team can trigger. It carries the context, not a copy of your inbox.
Switch models without moving anything
Your context, agents, and skills live in the workspace. Pick whatever shipped this week and nothing needs rebuilding.
Private until you say otherwise
Nothing you write is visible to anyone else until you share it. No bot sits in your inbox reading everything.
How it works
Give it a job, not a prompt.
Describe the work in a sentence. The agent runs with your context, on the model you chose, and shows what it did as it does it. When it reaches something it should not decide alone, it stops and asks.
A run in progress
Put it to work
Every team has a job worth handing over.
Reconcile what was billed against what was agreed.
It reads the contracts, pulls the ledger, and comes back with the variance and the four vendors behind it. Raising the credit memo waits for you.
Finance: the variance report
Keep the pipeline honest without chasing anyone.
It checks what moved since Friday, finds the deals with no next step, and drafts the nudge. You approve the ones that go out.
Revenue ops: the pipeline check
Answer the queue with the context your team already has.
It reads the thread, finds the last time this happened, and writes the reply in your words. Anything touching a refund stops and asks.
Support: a drafted reply
Run onboarding the same way every time.
Accounts requested, tasks created, intros booked, laptop ordered. It reports what it did and flags what it could not finish.
People: an onboarding run
Security
The boundary is enforced, not promised.
The reason companies will not put a bot in Slack is that it reads everything. Here, what an agent can reach is decided before the model is asked, and what it can change is decided before the call is made.
- Per conversation
- Access is decided by participant, not by a model asked to be discreet.
- Every write
- Money, records, and outbound messages are held for approval with the amount shown.
- Whole run
- Every tool call and every decision is written down and readable afterwards.
Pricing
Priced per person, with no charge for the agents themselves.
Team
$40/ person / month
Start a teamIncludes:- Unlimited agents and skills
- Your own context, private by default
- Any model, switched at any time
- Shared team skills and connections
Company
$65/ person / month
Talk to usEverything in Team, plus:- Roles, approvals, and an audit trail
- Connections scoped per team
- Single sign-on and directory sync
- Priority support
Regulated
Custom
Contact salesEverything in Company, plus:- Runs in your own cloud
- Approval on every outbound write
- Data residency and retention controls
- Security review support
The moment it becomes part of the team.
Placeholder quotes, pending real onesA short, specific sentence about the one job this replaced, in the customer's own words.
A second quote, shorter, about something the team stopped having to do by hand.
A third quote, comparing this against whatever they were using before it.
FAQ
Can an agent see something its person cannot?
No. Access is decided per conversation, by participant, before the model is asked anything. An agent is handed the context its person is entitled to and never the rest, so there is no case where the boundary depends on a model choosing to be discreet.
What stops an agent doing something expensive by mistake?
Anything that moves money, changes a record, or sends a message outside the company is held for approval. The card shows the amount and the recipient rather than just a tool name, so the person approving can see what they are agreeing to.
What happens when we want to change model?
Nothing moves. Context, agents, and skills belong to the workspace, not to a model provider, so switching is a setting rather than a migration.
Do we need an engineer to set this up?
No. Connecting a tool, writing a skill, and giving an agent a job are things an operations lead can do in an afternoon. Engineering is needed only if you want it running inside your own cloud.
Where does our data actually live?
In your workspace. On the regulated plan it runs in your own cloud with your retention and residency rules, and the audit trail is yours to export.
Meet your first agent.
We are onboarding a small number of teams by hand so the setup matches how you already work. Tell us what the first job would be.