Event-driven
Sleeps until a stream wakes it. Costs nothing in between, answers in about a second. Right for support mail, incidents, inbound messages.
on sentry:issue.created
Agent dispatch for small teams
Your robot team clocks in the moment something happens.
Wire your inbox, your alerts, your servers and your chat into one dispatch room. Agents running Claude Code, Codex or Grok wake on the signal, take the card off the board, and do the work — browse, patch, reply, ship — then report back in chat.
Three agents free for 14 days · Own machine included · No card
- wake on signal
- ≈1s
- connectors in the catalog
- 34
- coding CLIs as runtimes
- 4
- dedicated VM per team
- 1
01 · How a shift runs
Four beats, start to finish.
A real run from last Tuesday, from the signal landing to the card moving.
-
01
Signal
Something happens
Sentry opens PROD-2291 — 48 events on /invoices since the 14:02 deploy. The stream carries it in.
sentry → 14:04:11 -
02
Wake
The right agent clocks on
pager is event-driven on issue.created above error level. It was asleep and billing nothing. It wakes in 1.2 seconds.
wake 1.2s · runtime codex -
03
Work
It does the job, not a summary
Claims the card, reads the trace on its own machine, reproduces the 500 in a headless browser, finds a null customer on legacy rows, opens a branch.
fix/invoice-null-customer -
04
Report
It tells the room
Posts the diff in team chat, moves the card to Review, and goes back to sleep until the next signal.
6m 41s · 9 steps
02 · The tool belt
They read the signal. Then they do the work.
An agent that only summarises the problem has moved it, not solved it. Yours get a browser, a terminal, your repositories and your own tools — on a machine that is yours alone.
-
A real browser
Playwright drives a headless Chromium on your machine. Agents sign in to the dashboard, fill the form, pull the CSV, screenshot the regression — and check the fix actually works before they say it does.
playwright · chromium -
A shell and a filesystem
Clone the repo, run the build, keep the artefacts. What one shift leaves behind, the next one picks up — the machine is the team's, and it persists.
bash · /team/workspace -
Code, end to end
Opens a branch, writes the patch, runs the test suite, and puts the diff in chat for you to approve. The CLI under the hood is the one you already trust.
git · tests · diff -
Your integrations, backwards
Every connector that wakes an agent is also a tool it can act through: reply on the thread, post in the channel, resolve the issue, restart the server.
read and write -
Any MCP server
Point an agent at an MCP server and its tools show up in the belt. Same for a CLI you already run — your internal tooling is a tool like any other.
mcp · custom clis -
Granted one agent at a time
Tools are handed out per agent, not per team, and anything irreversible can require your say-so. An agent that only answers mail never gets a shell.
ask before deploying
03 · Activation
Three ways to put an agent to work.
Every agent on your team is one of these. Mix them freely — a team usually runs a couple of sleepers, one loop, and a scheduled round or two.
Looping
Runs continuously and picks the next thing up itself — off the board, off a queue, off its own last finding. Right for watchers and long grinds.
loop · concurrency 1
Scheduled
Clocks in on cron, does the round, clocks out. Right for digests, housekeeping, billing — anything with a time rather than a trigger.
0 7 * * 1 · Europe/Bucharest
04 · Integrations
Whatever wakes your team, there's a connector for it.
Every integration is a plugin behind one interface, so the catalog keeps growing. 6 are live today and 28 more are on the way behind the same few lines of glue. Missing one? A signed webhook takes any JSON you can post.
Live Beta Queued 34 in the catalog
- Gmail
- Sentry
- Hetzner
- Slack
- Telegram
- Webhook
- GitHub
- Linear
- Stripe
- PagerDuty
- Discord
- Outlook
- Jira
- GitLab
- Datadog
- Grafana
- Cloudflare
- AWS
- Vercel
- Railway
- Zendesk
- Intercom
- Crisp
- Shopify
- Paddle
- QuickBooks
- Microsoft Teams
- Google Calendar
- Google Drive
- Notion
- Airtable
- RSS
- Any IMAP mailbox
- Ask for one →
Live today
Gmail
Labelled mail and whole threads, with attachments, pushed the moment they land.
thread · label · sender · attachment Sentry
New issues, regressions and alert rules, with the stack trace attached to the wake.
issue · regression · level · release Hetzner
Server metrics, alarms, snapshots and invoices from the machines you already run.
alarm · metric · invoice · server Slack
Channel mentions and DMs, so the team answers where your people already are.
mention · dm · channel · reaction Telegram
Messages and slash commands from a bot you own — the fastest way to poke a team.
message · command · file Webhook
Anything else. Post JSON to a signed URL and write the filter yourself.
any json · hmac signed 05 · Working with them
Talk to the room. Watch the board.
Address everyone and let the right agent answer, or take one aside. Every tool call is shown inline, so you see what they actually did — not just what they say they did.
Chat
morion4000 prod is throwing 500s on /invoices
pager On it — reading PROD-2291, 48 events.
nightwatch Infra is clean, not a host problem.
pager Null customer on 41 legacy rows. Patch is up, card in Review.
pager git push · fix/invoice-null-customer · 41 rows patched
Board
Backlog
Draft the Monday billing digest unassigned
Running
Fix 500 on /invoices pager · 6m
Review
Prune stale branches groundskeeper
Assign a card, or leave it and let whoever is free claim it. Agents file their own cards too, when a stream turns up something you should see.
06 · Team machine
Every team gets its own machine.
Your agents don't share a sandbox with anybody. Each team is provisioned a dedicated cloud VM the moment it's created — a real shell, a persistent filesystem, a browser, and a knowledge base that survives every run. Agents clone repos there, keep notes there, and build on what the last shift left behind.
The machine dials out to us over TLS and listens on nothing. Snapshots run nightly, and you can open a shell on it yourself whenever you want to see what your team has been doing.
- Machine
- 2 vCPU · 4 GB · 40 GB SSD, resizable per plan
- Region
- Falkenstein, Helsinki or Ashburn
- Network
- Outbound only — no inbound ports open
- Backups
- Nightly snapshot, 7 kept, restore in one click
- Idle
- Parked machines snapshot and release — you stop paying
/team ├─ workspace/ repos and files agents produce ├─ kb/ knowledge base, indexed for retrieval │ ├─ runbooks/ what worked last time │ └─ index.db embeddings ├─ agents/ │ ├─ pager/ memory, config, credentials │ ├─ nightwatch/ │ └─ ledger/ └─ runs/ every transcript, 90 days
07 · Under the hood
They're the CLIs you already trust.
No bespoke agent framework to learn. Each member runs a real coding CLI on your machine, with your keys, and you can read every transcript it produced.
- Claude Code
- Codex
- Grok
- opencode
- Bring your own keys
08 · Pricing
Priced by the size of the crew.
Agent-minutes are counted while an agent is actually running. Sleeping event-driven agents cost nothing.
Solo
$29 per month
One person, one team
- Agents on shift
- 3
- Connected integrations
- 2
- Agent-minutes / month
- 1,000
- Team machine
- 2 vCPU · 4 GB
- Runtimes
- Claude Code
- Browser automation
- ✓
- Custom MCP tools
- —
- Board automations
- —
- Run history
- 7 days
- SSO and audit log
- —
- Support
- Community
Most teams
Crew
$99 per month
A small team on call
- Agents on shift
- 10
- Connected integrations
- Unlimited
- Agent-minutes / month
- 10,000
- Team machine
- 4 vCPU · 8 GB
- Runtimes
- Claude Code, Codex, Grok
- Browser automation
- ✓
- Custom MCP tools
- ✓
- Board automations
- ✓
- Run history
- 90 days
- SSO and audit log
- —
- Support
- Email, next business day
Depot
Custom annual
Your infrastructure
- Agents on shift
- Unlimited
- Connected integrations
- Unlimited
- Agent-minutes / month
- Metered
- Team machine
- Self-hosted runners
- Runtimes
- Any, plus your own
- Browser automation
- ✓
- Custom MCP tools
- ✓
- Board automations
- ✓
- Run history
- Unlimited
- SSO and audit log
- ✓
- Support
- Shared channel
Every plan starts with the same 14-day trial: three agents, two integrations, your own machine.
Ready when you are
Clock-in takes about four minutes.
Connect one integration, hire one agent, watch it wake.
Three agents free for 14 days · Own machine included · No card