Governed agent control plane
Your business ran while you slept.
Hire AI coworkers with approvals, budgets, and an audit trail — so overnight work doesn’t become overnight risk.
Not another chat UI. A governed control plane for agents that run on a schedule and stop for exceptions.
- Runs
- On a schedule
- Asks you
- Only for exceptions
- Keeps
- Health, outcomes, spend
Revenue Analyst
Last night, while you slept
Working…
An example of one morning’s hand-over.
A morning on the desk
Five minutes to a governed outcome
This is what ships today — not a pitch deck. Connect a tool, hire a role, clear what it may do alone, and open to exceptions.
Read the first-outcome guide- 01
Connect the tools you already use
GitHub for PRs, analytics for a weekly brief, mail for drafts that wait for your yes.
- 02
Hire a coworker from a role template
PR reviewer, morning chief of staff, on-call summarizer — prompt and capabilities start filled in.
- 03
Clear what may run alone
Pre-authorise routine actions. Anything outside the list holds in Inbox until a person stamps it.
- 04
Give it a shift
Recurring cron, one scheduled run, or straight away — then watch the session or wait for the morning digest.
- 05
Open to health, outcomes, and spend
Desk health, shipped vs blocked, budget ratio, and a weekly control report — not a chat scroll.
The staff
Hire an agent into a role
An agent is a job description with tools attached: what it knows, what it may touch, what it costs. Hire from a role template on Teams, give it a shift, and it turns up to work. Live presence shows who is on shift.
Scenarios by role- Morning desk brief
- Chief of staff role — overnight outcomes, decisions waiting, and spend vs budget before you open the desk.
- PR reviewer
- Reads the diff, flags risk, and leaves an approve / request-changes note under your guardrails.
- On-call summarizer
- Turns overnight incidents into a calm hand-over: impact, status, and open decisions.
- Research analyst
- Answers grounded questions from connected apps, cites sources, and files a short memo.
Weekdays, 07:00
On every draft PR
Daily, 07:00
Mondays, 08:00
The shift
Work that turns up on its own
A routine is an agent, a job, and when it runs. Once a week, once a month, or the moment something breaks — no one has to remember it.
Recurring
Every Monday at 07:00. Every 30 days. First of the month. It keeps its own calendar — work that turns up while you sleep.
Scheduled
One run, at a time you pick, then it stands down. Good for a launch or a deadline.
Straight away
Queue it now and watch it work. The same job, run by hand when you need it.
The point is not that it can do the job. It is that nobody has to ask it to.
Trust is the product
What it can do alone — and what always waits for you
Overnight value only works if risk stays contained. Approvals, budgets, denied commands, and a kill switch are first-class — not settings buried under a chat prompt.
Read how security worksMay run alone
- Actions you pre-authorised on the routine
- Read-only work inside connected apps
- Drafts parked in docs for morning review
- Spend inside the monthly caps you set
Always waits
- Anything outside the allowlist — Inbox stamp required
- Denied patterns (force push, rm -rf, and your list)
- Sending mail, money moves, or production-touching writes you did not clear
- Runs that would breach budget — paused until you raise the cap
Pre-authorise the routine
Tick what a job may do on its own — send the mail, write the doc, file the ticket. It stops being a question and starts being the work.
Spend confidence before it runs
Every agent and project carries a monthly cap. The desk shows budget ratio and an estimated next-run cost when history exists — it stops before it blows the number.
Every shift is on the books
What it read, what it changed, what it sent, what it held back — with outcome memory (shipped vs blocked) and a weekly control report.
Evidence on the desk
What you can verify without a sales call
We do not invent customer quotes. These are product facts you can open after signup — or ask us to walk through on a demo.
- Outcome memory
- Shipped vs blocked across runs — so the desk remembers what actually landed.
- Weekly control report
- Approvals, spend, and pauses in one operator-facing summary (email when you turn digests on).
- Your keys, your repos
- Model spend stays on your provider keys. Credentials are encrypted; the model never sees the raw secret.
- Design partners welcome
- If you want a named case study later, we earn it on a real shift — start with Contact and the first job.
It works where your business already runs
Browse the directoryPricing
Priced like a desk, not a seat
You buy control-plane capacity — seats, agents, apps, concurrent runs — not marked-up tokens. Hire AI coworkers with approvals, budgets, and an audit trail — so overnight work doesn’t become overnight risk.
Prefer to prove it first? Start free — 1 agent, 20 Chat conversations, then upgrade when the desk needs room.
Starter
1–2 ops owners proving the desk on a single project.
$50
per month
A small team, one desk, and agents that stop when you say so.
Start with StarterGrowth
Most popular
Multi-project eng / ops desks that need sandbox and the full app directory.
$100
per month
More seats, sandboxed shell, custom MCP, and the full app directory.
Choose GrowthScale
Larger delivery surfaces running many agents in parallel under shared support.
$300
per month
Room to run many agents in parallel across a larger delivery surface.
Book a walkthroughQuestions
What people ask before they hand something over
- Does it really run without me?
- For the work you cleared, yes. You tick which actions a job may take on its own when you set it up. Anything you did not clear waits, and the run picks up the moment you decide. A morning desk digest can land in email so you open already briefed.
- What stops it doing something stupid at 3am?
- Three things: it can only touch apps you connected, only take actions you pre-authorised, and only spend inside a monthly cap. Desk health and the weekly control report show approvals, spend, and pauses. Anything outside that holds until morning.
- How is this different from a chatbot?
- Not another chat UI. A governed control plane for agents that run on a schedule and stop for exceptions.
- What do I have to set up?
- Connect the tools you already use, hire an agent from a role template (or blank), and give it a shift. The first useful job is usually a governed review or a weekly brief — we guide that first outcome on the desk.
Give it the job you keep forgetting to do
Tell us your role and the first shift you’d hand over. If Okestri can run it under approvals, we’ll show you it running — or point you at Free to prove it yourself.