An answer with its evidence
Each answer carries up to 12 evidence chips. A trace chip opens the trace, so you can check the claim instead of trusting it.
Platform
Ask a question about your own telemetry.
Ask Onyx Oracle opens from the top bar. Ask why a service is failing and it reads your traces, logs and metrics, then answers with a summary, findings, hypotheses, next steps and the evidence it used.
What you get
Limitations next to the answer
The answer lists what it could not establish. Treat its hypotheses as leads to check, not conclusions.
Actions need a person
On an investigation Onyx opened for a critical incident, Oracle can propose up to four actions: acknowledge the incident, pause the rule, add a note, or run an existing workflow. Nothing runs until an admin approves it.
How it works
Ask in plain words
Type a question of up to 8,000 characters in the top bar.
It reads your workspace
The request goes through a signed relay tied to your workspace. API tokens cannot ask questions.
You get the answer in sections
A summary, findings, hypotheses, next steps, limitations and evidence.
Limits worth knowing
- Per-person limit. By default each person can start four questions every 30 minutes in a workspace.
- No history. The conversation is not kept after you reload the page.
- The time-range picker is not sent. Say the period you care about in the question.
- A third-party model. Questions are processed by an AI model provider that our Privacy Policy lists.
Ask it something you would ask a teammate.
Trial workspaces include Onyx Oracle.