In use at
Point at the spike. Everyone asks why.
A bot keeps the reliability dashboard. Someone circles Tuesday 4pm. One person knows a deploy went out then, one knows a pipeline ran late, one saw the same alert fire last week. Each points to where the data lives. The bot connects to all three, tests each theory against the spike and adds a panel with the answer.
Start a dash bot →
Comment on the bad cell. It gets fixed.
A bot owns the data quality sheet. One person knows the upstream schema changed, one knows an event is double-counted on purpose, one knows which source wins when two disagree. None of that is in the data. The bot fixes each at the source and rebuilds the sheet. Every fix is logged.
Start a sheet bot →
Review the spec together. Decide it in the doc.
A bot maintains the spec. One engineer knows two customers already use the old API, one knows why the last migration failed, the lead knows what was promised for the quarter. The bot checks each against the code and the tickets and writes the decision into the doc, next to the comment.
Start a doc bot →
The metrics review, built by everyone who knows the numbers.
A bot drafts the metrics review from the warehouse, the dashboards and last month's deck. One person knows the pipeline double-counted for a week, one knows the definition of "active" changed mid-quarter, one knows the spike is a migration, not growth. The bot rewrites the slides around what they know. Nobody has to build the deck.
Start a deck bot →
Leave the meeting with the plan already on the board.
A bot runs the board. Paste in the standup notes. One engineer knows the vendor is late, one knows a review is blocked on someone's leave, the lead knows the migration window moved. None of it is written down yet. The bot reworks the plan around all three and tells the people it affects.
Start a board bot →
One bot.
- 01A bot can run 24×7.
- 02A bot can work on its own isolated computer, with a browser.
- 03A bot can build tools to connect to any API or database.
- 04A bot can create artifacts and apps.
Many users.
- 01Users can't ask the bot to access another user's data.*
- 02Users can edit and delete messages in the shared chat.
- 03Users can spawn private bots to do secret work and report back.
Start a bot. Invite a collaborator.
Spin up a bot and give it a responsibility or a goal. When it hits a wall, or has output worth sharing, bring someone in and keep things moving.
Real work, by team
Board deck
CEO · Finance EngineeringDraft it from last quarter's. Tell me what data you're missing and who you need to talk to for the narrative.
Competition radar
CEO · ProductInfer from my sent emails and our deal notes which competitors I should track. Suggest who else should add names. Then run a twice-daily digest for the team.
Ad-hoc dashboard
DataSomeone asked for a dashboard. Build the basic version from our existing dashboards, then tag them to check it's what they actually wanted.
Dedupe metrics & dashboards
DataWe have overlapping and conflicting dashboards and metric definitions. Find them and bring in the right folks so we can document which ones are correct.
Stale pipeline
Data · EngineeringThe orders model hasn't refreshed since yesterday. Trace the pipeline, find the broken dependency, pull in whoever owns it, and post the fix with evidence.
Actionable bug from a customer report
EngineeringReproduce the issue or read our logs, find it in the code. Pull in the right engineer when stuck. Flag if it needs a product call.
On-call
EngineeringAlert in a service I don't know. Pull recent changes, similar incidents. Identify the right expert if stuck. Send them your hypothesis and what to confirm.
Access review
Security · DataWho can reach the customer tables, and which services use them? Flag anything that looks wrong and draft the fix. Don't change access without my approval.
Security questionnaire
Security · GTM EngineeringSplit the questions by who'd know: infra, controls, DPA. Reuse past answers; bring people in only for the rest.
Communicate a deliverable ETA
Product · EngineeringWhen does [issue] ship? Is it in a sprint, backlogged, or off the radar? Ask the owning engineer. Check what we've already told them. Draft my reply.
Clickable prototype
Product · EngineeringTurn this spec into wireframes I can click through. Get design to mark what's off and engineering to flag what's expensive. Iterate until it's decided.
Feature adoption
Product · DataAdoption for [feature] disagrees between two event streams. Find out why, fix the instrumentation, and give me a number we can trust.
ARR reconciliation
Finance Engineering · DataBilling, CRM and the warehouse each say a different ARR. Trace every number to its source, find the join that's wrong, document which one wins.
Flux analysis
Finance EngineeringBuild it from the ledger with last month's as the baseline. Where a line needs a story, pull in whoever owns it.
Attribution
GTM Engineering · DataCampaign attribution is off. Trace it from source events to the warehouse, find the missing UTMs and the dupes, add checks so it stays fixed.
Account brief
GTM Engineering · ProductCall with [account] Thursday. Pull usage, open tickets, and what they asked for last time. Get product to confirm shipped vs roadmap vs never.
Website edits
GTM EngineeringFix the copy on [page]. Show me the diff, get product to confirm the claims, open the PR. Nudge me when it's live.
People data
HR Engineering · SecurityWhich tables hold sensitive people data, who can read them, and should they? Build a masked dataset for the analyses we've approved.
Onboarding
HR EngineeringNew engineer starts Monday. Work through the checklist: accounts, repos, laptop, first-week buddy. Ping the owner for anything that needs a human approval.
Pay only for the AI you use
Every user starts free. Pay only for users who need more in $40 increments, billed weekly.
- Shared context: knowledge, skills, semantic layer
Consumption based with a monthly minimum commit.
- No limit on users
- Shared context: knowledge, skills, semantic layer
- Single-tenant deployment
- BYOC or self-hosted deployment
- Bring your own models
- Dedicated forward-deployed engineers
What is a bot, exactly?
An AI coworker you spin up for one thing: the pipeline that broke, the API migration, the dashboard nobody trusts. It starts generic and becomes yours as you use it. What it learns, what it makes and what you told it all stay with it.
Can my teammates review the same work?
Yes. Invite them into the same bot. They see the same work, correct the assumptions they know are wrong, and the bot keeps going.
What can it see?
Only the data and integrations available to the person who triggered it. A teammate's credentials don't become yours. In a shared bot, everyone who can see the bot can see what it shares, so use a private bot for sensitive work. How that works →
Whose permissions does a bot use?
A bot has no access of its own. Each time it acts, it borrows the permissions of the person who asked: the data, apps and files that person can already reach, nothing more. If you ask it for a report, it can only pull from what you can see. If a colleague asks the same bot next, it switches to what they can see.
So a shared bot is not a side door. You can't ask it to fetch something a teammate can see and you can't, and neither can they. Access is decided by your organisation's permissions, not by who happens to be in the chat.
One thing to keep in mind: whatever a bot says in a shared chat is visible to everyone in that chat. If the question is sensitive, ask it in a private bot and share the result.
Where does it run?
On its own computer, with its own browser, around the clock. Nothing is installed on your phone or laptop, and it keeps working when you close the app.
Can I work on a sensitive investigation privately?
Yes. Start a private bot for the investigation and share only the reviewed findings with the wider team.