Ask your releases
anything.
An assistant that reads your flags, traces and dashboards. Ask why something broke and get the answer with the chart, the commit and the rollback.
Reads only what you connect · Never trains on your data
Why did checkout get slower at 14:02?
new-checkout moved to 25% at 14:01. p95 latency went from 210 ms to 910 ms, and the rollback rule dialled it back to 0% at 14:03. Nobody was paged.
Dashboard · checkout latencyWhat changed in that release?
Commit 8f3c21 swapped the tax service client. The new client retries three times on timeouts, which tripled calls at peak.
GitHub · PR #2291
- 4.2 s
- median time to an answer
- 93%
- answers with a linked source
- 31 min
- saved per incident review
- 0
- bytes used for training
Grounded in your
own releases.
The assistant only answers from what it can see: flag history, traces, dashboards and pull requests. If it cannot find a source, it says so.
Connect
Link your workspace, repositories and metrics. Read-only, scoped per project.
connect githubAsk
Plain questions, in chat, Slack or the command palette. No query language.
/ask whyAct
Every answer offers the next step: dial back, open the trace or page the owner.
dial 0%Questions teams ask every day.
Incident review
“What happened between 14:00 and 14:10?” Timeline, cause and fix in one answer.
Release notes
“Summarise what shipped this week.” Written from merged pull requests and flags.
Agent health
“Is the support agent drifting?” Judge scores, cost and failure patterns.
Flag clean-up
“Which flags are safe to delete?” Stale flags with their owners and last change.
Experiment readout
“Did the new pricing page win?” The verdict, the lift and the confidence.
On-call handover
“What should the next shift know?” Open rollouts, alerts and risky changes.
Every claim
has a link.
Answers cite the chart, trace or commit they came from. One click takes you to the evidence, so nobody has to take the assistant’s word for it.
Summarise what shipped this week.
Four releases went to 100%: the Go SDK speed-up, warehouse export, the MCP server and judge-based rollbacks. One rollout, new-checkout, was rolled back on Tuesday.
Changelog · 4 releases
Reads the tools
you already use.
Read-only connections, scoped per project and revocable at any time.
- GitHub
- Slack
- Datadog
- OpenTelemetry
- PagerDuty
- Linear
- Jira
- Snowflake
- Rheostat traces
See it answer
a real incident.
Three minutes, one outage: the question, the answer with its sources and the rollback, straight from the chat.
The teammate who read every log.
“Our incident reviews start with the assistant’s timeline now. It is right more often than our memory is.”
“Release notes used to take me an afternoon. Now I edit a draft for ten minutes.”
“The links are the point. I never have to trust an answer I cannot check.”
Assistant questions.
- Which model does it use?
- A mix of hosted models behind our own retrieval layer. Enterprise plans can bring their own model endpoint.
- Is my data used for training?
- Never. Your data is used to answer your questions and nothing else, and it stays in your region.
- Can it change things?
- Only with a confirmation. Dialling back a flag or paging an owner always asks you first.
- What does it cost?
- Included in Team and Enterprise with 2,000 questions a month; more can be added in blocks.
Ask your first
question today.
Connect one repository and one dashboard. The assistant is ready before your coffee is.