AI Assistant

The AI Platform Assistant helps teams investigate production behavior in natural language. It answers questions with evidence from selected telemetry sources, not generic guesses.

What The AI Assistant Does

The assistant can query traces, evals, dashboards, cost signals, model analytics, tool usage, process units, and business events. It can summarize incidents, compare time windows, find likely root cause, and recommend next steps.

When To Use It

Use it when you need to:

Explain a spike, regression, or cost increase.
Summarize an incident for a stakeholder.
Compare models, tools, workflows, or releases.
Generate a query or dashboard starting point.
Investigate without learning a new DSL.

Prerequisites

Ingest the telemetry sources you want the assistant to use.
Add useful dimensions like workflow, user, team, model, deployment, and business outcome.
Select allowed sources before asking a question.

Step-By-Step Setup

Open AI Assistant.
Select data sources such as token usage, model latency, tool usage, evals, or cost.
Ask a specific operational question.
Review evidence links and source records.
Save the answer as an incident note, report, dashboard idea, or follow-up task.

Example Questions

Why did checkout-agent latency spike yesterday?
Which model caused the cost increase this week?
Show failed tool calls for support-agent grouped by root cause.
Summarize quality regressions after the last deployment.

What Appears In The Console

Answers include selected sources, record counts, supporting traces or dashboards, confidence context, and generated summaries that can be shared with engineering, product, support, or leadership.

Common Mistakes

Asking broad questions without selecting relevant sources.
Treating generated summaries as proof without opening supporting evidence.
Forgetting to include business fields in telemetry.
Giving the assistant access to sensitive data that should be redacted.