Nova · Your AI SRE
Find the cause.Fix it safely.
Nova investigates production issues with Astro Platform evidence and connected tools, proposes fixes for your approval and helps verify recovery. Sensitive reads ask permission first.
Explore AI SRE examplesNova is free to use with approved access. Usage limits apply.
Nova · incident investigation
pods-probe-web
Crash loop · production incident
Production issue detected
Nova gathers events, rollout history and permitted evidence
Based on Nova's recorded INC-79 investigation · human approval required for changes
Who it is for
You are on call for something you did not build all of.
- The first twenty minutes of an incident go to working out what changed.
- You want it to read production for you, not change it unsupervised.
- Someone will ask afterwards what was run, and who approved it.
One incident · INC-79
You · 23:51 PDT
Why does pods-probe-web keep crash-looping?
Nova · evidence after log consent
chown("/var/cache/nginx/client_temp", 101) failed
Operation not permitted · App logs
Your decision · 23:55 PDT
Approve the nginx-unprivileged image.
The recorded change runs after approval. Health checks follow.
Inside a real investigation · INC-79
Four steps. Two of them are yours.
See Nova’s AI SRE work through a crash-looping app. Start with the finished finding in chat, then follow the evidence, approved change, recovery checks, and postmortem behind it.
The finished finding, in chat
One question. A finding you can inspect.
“pods-probe-web keeps crash-looping in production. Why?” The chat brings the finding and its source together. Open the full record to follow the evidence, change, and decisions.
123:52–23:53 PDT
Read the evidence, with permission.
You allow app logs. Nova finds the startup error: nginx cannot change ownership of its cache directory while running as a non-root user.
INC-79 · EvidenceOpen zoom view 223:54–23:55 PDT
Review the change before it runs.
Nova proposes the nginx-unprivileged image, with steps, risk, and an undo. You approve the redeploy. The record shows the fix was applied.
INC-79 · FixOpen zoom view 3After the approved change
Check recovery. Keep the decisions.
Health checks later report RUNNING with no warnings in the last five minutes. Consent, approval, and follow-up checks stay in the activity record.
INC-79 · ActivityOpen zoom view 4Postmortem · INC-79
A postmortem the next person can use.
The postmortem connects the cause, approved action, and outcome. It also states that customer impact could not be confirmed from the available telemetry.
INC-79 · PostmortemOpen zoom view
Worked examples
Start from a real incident.
Each example has its full walkthrough in the docs.
- Investigate a failing appThe cause, with the line that proves it, and what it could not check.Read the walkthrough
- An alert that starts its own investigationNova posts in your incident channel and starts reading.Read the walkthrough
- Restore service first, then approve the fixWhen a deploy lines up, the rollback comes before the cause is confirmed.Read the walkthrough
- The postmortem, draftedBlameless, from the record. Publish it to GitHub, Slack, or both.Read the walkthrough
- Work an incident from Slack or the terminalFollow it and answer it where you already are, without opening the console.Read the walkthrough
Trust and governance
Review the fix before Nova applies it.
In INC-79, you allowed the logs and approved the image change. The runtime keeps those decisions separate.
You approve the exact change
Only the change shown in the card runs.
Sensitive reads ask first
Choose Allow once or Allow for this chat for the same app and tool. A changed tool schema asks again.
Health checked after every change
If the app is not healthy, it reads Fix failed.
Investigation record
Evidence and decisions from this incident stay together.
Your organization only
Evidence and what Nova learns stay inside it.
Trust earned per server
Every call from a new server asks a person until an admin trusts it.
Explore the guardrail policy
Your rules. The runtime enforces them.
Every organization starts with safe defaults. An admin changes any row. The model never sees a blocked action.
Reads
health, events, metrics, cost
AllowedRequires approvalBlockedMessaging
Slack updates and channels
AllowedRequires approvalBlockedSensitive reads
application logs, security scans
AllowedRequires approvalBlockedRemediation
restart, rollback, deploy, update
AllowedRequires approvalBlockedScaling
replicas and resizes
AllowedRequires approvalBlockedInfrastructure changes
provisioning
AllowedRequires approvalBlockedDeletion
clusters and applications
AllowedRequires approvalBlockedSecrets
credentials, tokens, keys
AllowedRequires approvalBlockedInfrastructure teardown
destroying what was provisioned
AllowedRequires approvalBlocked
Shadow mode
Try a policy on reads first; writes stay enforced.
Emergency stop
One switch halts every write, for the whole organization.
Your MCP tools
Same policy; every call asks a person until an admin trusts the server.
You can reject a proposal, fix the app yourself, or stop the investigation. Stop prevents further work in that run; it does not undo changes already applied.
Works with
It reads what you already run.
Astro Platform gives Nova direct access to the apps, clusters, deployment history, and health it manages. Your connected tools add evidence from the rest of your environment.
- Context from production
- Nova reads platform state alongside logs and events. In INC-79, it connected the startup error to the app’s non-root configuration.
- A path to the fix
- With your approval, Nova runs supported changes through Astro Platform. The proposed image change became an applied redeploy.
- An outcome you can check
- Nova checks the app’s health after the change and preserves the result with the investigation. You can inspect the evidence and decisions.
Astro Platform remains available through its Console, CLI and API. Nova is the engineering companion that works across it. Explore Astro Platform →
- Astro PlatformEnterprise
- GitHubEnterprise
- AlertmanagerEnterprise
- GrafanaEnterprise
- SlackEnterprise
- DatadogEnterprise
- Your MCP serversEnterprise
- Your runbooksEnterprise
Plans
Free runs Nova. Enterprise scales it.
Every plan has the full Nova, limited by units. Enterprise adds more units, Slack and your own tools.
Developer
Find the cause and fix it, within your units.
- Standard and deep investigations
- Built-in knowledge of how apps fail
- The cause, with its evidence quoted
- Nova applies the fix you approve, and checks it worked
- Learns from past incidents
- On-call from the CLI, and you choose who approves
- Share with a link
- Alerts start deep investigations, grouped, with results in the console
- 175 units a month, 20 a day
Enterprise
Everything in Developer, with more units and your team's tools.
- More units, pooled for the organization
- Alert results posted to your Slack channel
- On-call in Slack
- Learns from your runbooks through Operational Memory
- 90-day audit trail
- Your MCP servers and Datadog
Before you ask
The questions that come next.
Can it change production on its own?
Supported Nova remediation workflows require a person to review and approve the proposed action before it runs. Nova then helps verify recovery where a health signal is available. Other connected tools retain their own permissions and controls.
What does it read, and who sees it?
Events, rollout history, configuration, metrics and cost as first-party state, plus what your other connected tools can read. Application logs and security findings ask first. Allow once covers one read; Allow for this chat covers the same app, tool and connection until the grant expires. A tool schema change requires fresh consent.
Is our data used to train models?
Investigation reasoning runs on model providers whose commercial terms exclude training on your data. Some supporting tasks use endpoints with different terms, and our privacy policy lists which.
How is it priced?
Nova measures investigation work in units: 5 for a standard investigation or 10 for a deep one. The cost is shown before it starts. If Nova cannot read any data, the cost is 1 unit; once it reads data, the full investigation cost applies even if you stop it or no cause is found. An alert starts a deep investigation on every plan: results reach the console on Developer and your Slack channel on Enterprise. Access and limits depend on your plan.
Can I ask Nova questions before requesting access?
Public Ask Nova can answer questions about published AstroPulse information. Investigating or acting on connected infrastructure requires full Nova access for your organization, plus the applicable permissions and approvals.
What happens when it cannot find a cause, or is wrong?
It says so: Cause found, Likely cause or No cause found, with the evidence quoted and what it could not check. Review the evidence before accepting the diagnosis; a supported finding can still need further investigation.
Can I stop it or keep using the chat?
Yes. Stop cancels the investigation and prevents further work in that run. It does not undo a change that already ran. You can keep asking questions in the chat while an investigation runs; starting it again does not create an overlapping investigation.
What do we hand an auditor?
The investigation record documents evidence, proposed actions, approvals and what actually ran, including any verification gaps. Retention follows the applicable plan and organization settings; see our Privacy Policy for details.
AI SRE
Bring a real production question.
Bring a production question. See the evidence, review the proposed change and keep a record through recovery and postmortem.
WHAT YOU CAN EXPECT
- 01
Investigate with evidence
Follow actual observations, evaluate causes and preserve uncertainty where proof is missing.
- 02
Approve the exact change
Sensitive reads require consent; proposed remediation waits for human approval.
- 03
Verify and document recovery
Keep the health checks, decisions and postmortem together in the investigation record.