Agents that run your departments. Not drafts. Finished work.
Each department gets a team of AI agents that take work all the way to done. You set the guardrails once — they only stop to ask about spend over your limit, something they have not done before, or an account you flagged.
Activity
nothing waitingWork from every team, as it finishes. Most of it never stops here. An item only waits when a guardrail says a person should see it.
Demo company with example data, not customer numbers · open the demo workspace
A team of agents for every department.
Eight teams, one signup. Every team runs on the same company data, the same guardrails, and the same action log, so adding a department does not mean learning a second product.
Outbound that researches, writes, and sends.
The Sales team researches from your CRM history, past threads, and the live web. Then it writes the sequence, sends it from your connected Gmail or Outlook account, and books the meeting with prep notes.
- →Research attached to every send. Each claim links to the note it came from, so the log shows the reasoning, not just the output.
- →Replies answered the same hour. Automatic reply detection stops the sequence when the answer lands, and the agent answers back.
- →Meetings booked with prep notes. Conflict-checked invites on Google Calendar, CRM updated, prep ready.
Pipeline hygiene that fixes itself.
Watchers track stage age, next steps, and close dates across the pipeline, then write the fix instead of filing a report. Forecast rolls up from activity, and every override stays on the record. Salesforce and HubSpot are CSV import today, so the honest claim is that this runs on your CRM data, imported.
- →Watchers, not weekly reports. Stalled deals, missing next steps, and quarter-end pileups are caught as they happen.
- →Hygiene applied in bulk. Hundreds of field fixes a week, each one logged with the record it touched.
- →Numbers that tie out. Headline figures come from database aggregates, never from a model adding up rows.
| Deal | Amount | What it did |
|---|---|---|
| Crestline Payments | $92,000 | Close date moved three times |
| Alder Finance | $64,000 | Next step written from the call notes |
| Beacon Clearing | $118,000 | Champion left, new contact found |
| Juniper Capital | $45,000 | Contract nudge sent, opened twice |
| Pacific Ledger Co. | $71,500 | Single-threaded, second contact queued |
Content ships. Ad spend moves itself.
Marketing agents write and publish blog and social posts on the calendar, build campaigns, and report what worked. Ad spend reads are live across Meta, Google, and LinkedIn, and spend moves run inside your limits today. Write actions on a channel go live only after that platform approves them.
- →Published, not queued. Posts publish to LinkedIn, X, and your blog on the date the calendar says.
- →Spend rebalanced daily. Underperformers paused, winners funded, all inside your daily ceiling.
- →The big move asks first. Anything over your limit stops and names the person it asked, like the one on the right.
| Campaign | Budget / day | Cost per lead | What it did |
|---|---|---|---|
| Webinar promo | $380 | $71 | 6 leads, low volume |
| Brand search | $420 | $38 | Steady for 6 weeks |
| Retargeting · recent visitors | $640 | $54 | Left alone, inside its band |
| Prospecting · fintech CFOs Meta | $6,400 | $118 | Cost per lead up 41%, move drafted |
From issue to merged pull request.
Point an agent at an issue. It plans, builds in a branch, and opens the pull request on GitHub with tests green. Every merge needs your approval, and production touches always ask first.
See it on Engineering →#312 · Fix flaky session refresh in auth tests
agent/312-auth-flake → mainRoot cause: two tests shared one session fixture, so refresh raced token expiry. The fix isolates the fixture per test and adds a regression test for the race.
Customer Success
BetaRenewal risk, QBR prep, and account health from your own usage and support history.
See it on Success →Finance
BetaBudgets, invoices, and approvals reconciled against your own accounts, read-only to start.
See it on Finance →Support
BetaAnswers drafted from your docs, cases triaged and routed, help center kept current.
See it on Support →Legal
BetaContracts reviewed against your playbook, redlines drafted for counsel, renewals never missed.
See it on Legal →One system of record underneath.
Every team reads and writes the same pipelines, contacts, deals, and activity. Import contacts and history via CSV, or let automatic mailbox backfill build them from your Gmail. Gmail and Outlook send over OAuth inside your caps. It is why a RevOps fix shows up in the Sales agent's next email, and why the forecast does not need a second source.
Pipelines and deals
LiveDeals, stages, amounts, and close dates, with the forecast rolled up from activity rather than typed in.
Contacts and companies
LivePeople, accounts, and the whole thread of what was sent and what came back, which is what the agents read before they write.
Watchers and analytics
LiveWatchers that fire on a change in the data, plus first-party web traffic and multi-touch attribution.
You decide what runs on its own.
Write the rules once. Almost everything runs without you. The exceptions are the ones you would want to see anyway, and each one arrives with the reason it stopped.
- →Automatic by default. Outreach, replies, field fixes, published posts, and spend inside your bands all run without a person.
- →Limits, not inboxes. Set a daily spend ceiling, a volume cap, or a list of flagged accounts. Those are the only brakes.
- →New kinds of action pause once. The first time an agent tries something it has not done before, it asks. After that it is a rule.
- →Nothing is invisible. Automatic actions sit in the log next to the ones a person touched, with the rule that let them run.
Everything it did, and why.
One log across every team. Each row names the action, the agent, and the rule or person that allowed it, so an agent working on its own is still an agent you can audit.
Connect your tools
Gmail and Outlook mail over OAuth, paste-token GitHub, Slack, Notion, and Loops, and your CRM data by CSV import or automatic mailbox backfill. Teams share one company record, so Sales and Marketing are looking at the same accounts.
Set the guardrails
Spend ceilings, volume caps, flagged accounts, and who to ask when something crosses one. This is the only configuration that matters.
Agents do the work
They research, write, send, fix, publish, and rebalance, on their own, all day. Work that crosses a guardrail waits and says who it asked.
The log keeps the record
Every action timestamped with its reason, automatic ones included. Usage and productivity CSV export today; full action-log export is live on the Company plan.
Run it like the rest of your stack.
Access, spend, and audit controls are part of the platform. Here is what your IT review will ask about, and what is not built yet.
Roles and team seats
Owner, admin, member. Seats are scoped to a team, so Sales agents and Sales data stay with Sales.
Guardrail rules
Scope by team, action type, and amount. Require a second person above a limit. Rule edits are themselves logged.
Budget caps and alerts
Cap each team's monthly usage. You get an alert at 80 percent, and agents stop at the cap instead of running through it.
Action log
Export on CompanyEvery action with its reason and its evidence, on every plan. Usage and productivity CSV export today; full action-log export is live on the Company plan.
Single sign-on
In developmentSAML with your identity provider, so there is one place to offboard and one place to enforce MFA. Not shipped yet.
Usage you can read
Usage is metered monthly with a report per agent and per action, so finance can see what every dollar did.
From $750 per team per month, plus metered usage.
Turn on the teams you need. Each one is a flat license per month, and usage is metered on the same invoice. You set a budget cap per team, get an alert at 80 percent, and agents stop at the cap.