Alert source
Google Cloud
Errors from your logs, read through a service account you control. Read-only roles.
The AI agent for SRE that knows how your system breaks.
Corvo reads your telemetry, code and runbooks to find the root cause of each production alert, then leaves a draft pull request or an issue for your team to review.
Works withA sample workspace. Clear a row or open an investigation to try it.
The answer first. The steps when you want them.
Every investigation opens on its latest answer, and the Inbox lists only what needs a person.
Opens on the latest answer. Ask a follow-up and Corvo keeps working.
From a log line to an answer
Corvo holds alerts for five minutes, joins the lines of each stack trace, and hands every error to the investigation that already knows it. Each one ends with an answer, and a draft pull request or an issue if your team allows it.
Corvo remembers what it learns
Each investigation keeps its own notes. A fact for the whole workspace has a name and a history, and the next run reads it.
Connectors
Read-only access to your cloud, a GitHub app on the repos you choose, and the chat your on-call team already uses.
Code
GitHub
The Corvo app reads the repos you pick. Permissions decide whether it opens draft pull requests, issues or neither.
Alerts and chat
Telegram
Each alert is one message that updates in place while Corvo works.
Alerts and threads
Discord
A card per alert, with the progress and answer in its thread.
Tools
MCP servers
Add any MCP server, with OAuth sign-in for servers that need it.
Latest releases
All releasesv0.2.0
Team workspaces and alert investigations
Corvo now runs in a shared workspace with members and roles, and it groups alerts into investigations you can follow and steer.
Read morev0.1.0
GitHub App connections, PR replies, and shared team alerts
Corvo now connects through a GitHub App, replies inside pull request threads, and works in Telegram groups.
Read morev0.0.17
Incident deduplication with drop, skip, and process decisions
Corvo now records a drop, skip, or process decision for every repeated alert.
Read moreQuestions
Something else? Write to [email protected].
A Google Cloud service account with read-only roles, the Corvo GitHub App on the repositories you pick, and a Telegram or Discord channel for alerts. Setup lists every role before you connect.
No. Corvo reads logs and metrics and proposes changes as draft pull requests. Permissions decide which Git and GitHub actions it may take, and actions that publish or discard work need explicit trust.
No. You choose in Permissions whether Corvo may open draft pull requests, GitHub issues or both. With both off, Corvo still posts its answer and the evidence behind it.
Today, Google Cloud logs from Cloud Run, GKE and Compute Engine, GitHub for code, Telegram and Discord for alerts and chat, and any MCP server for other tools. More log sources are on the way as connectors.
It holds alerts for five minutes, then hands each error to the investigation that already knows it or starts a new one. The lines of one stack trace count as one alert.
Each investigation keeps notes on its root cause, its status and what to check next time. Facts for the whole workspace, such as where a service's logs are or how your team wants fixes written, each have a name and a history, and runs read them as soon as they are saved. If memory grows too long to read in full, the Memory page says which facts Corvo can't see. Corvo saves what it learns, your team can add, edit, delete or restore any fact, and only a member's preference changes how Corvo works.
You can follow every step while Corvo works. A run starts wrapping up at 50 minutes and stops at 55.
In an isolated sandbox for each investigation, prepared with your repositories and setup commands. One investigation can't touch another's files.
Yes. A workspace holds your connectors, investigations and reports. Invite teammates as Admin, Editor or Viewer.
Sleep through the next alert.
Connect your logs and your repositories, and Corvo takes the next one.