Trace setup

Send a captured agent run to the demo ingest API, or request an async review from sanitized trace metadata.

Async trace-shape review

Send a sanitized shape before wiring live traces.

Share your current trace/log shape, current workaround, and privacy constraints. Raw prompts and outputs are not required for the first review.

Request async review

Current trace/log shape

Send field names, event types, span examples, or a rough JSON outline.

Current workaround

Name the dashboard, spreadsheet, provider export, or custom logs you use today.

Privacy constraints

State whether metadata-only review, redaction, export-only, or self-hosting is required.

Demo API key

Builder plan
trq_dev_key

Use for the local demo only. Never commit production keys to version control.

Email configuration

Checking…

Debug a run — 4 steps

01

Create a trace

Send one agent run with a sanitized task summary and status.

02

Add spans

Attach LLM calls, retrieval steps, tool calls, retries, and errors.

03

Inspect the timeline

Open the run to find the failed, slow, or expensive span.

04

Tune capture

Keep raw prompts and outputs off by default unless explicitly enabled.

Current ingest examples

const response = await fetch("http://localhost:3000/api/v1/traces", {
  method: "POST",
  headers: {
    "Authorization": `Bearer ${process.env.TRACEIQ_DEV_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    name: "Support classifier run",
    agentName: "support-classifier",
    model: "anthropic:claude-sonnet-4.5",
    status: "success",
    input: "Sanitized one-line task summary",
    metadata: {
      captureInput: false,
      captureOutput: false,
      redactionStatus: "redacted",
    },
  }),
});

const trace = await response.json();

Capture defaults and controls

Captured by default

  • Trace name, agent name, status, and timestamps
  • Span name, type, status, duration, tokens, and declared cost
  • Retry counts, error messages, and suspect span metadata
  • Sanitized one-line input summaries

Disabled unless you choose it

  • Raw prompts and full completions
  • Customer records or tool payload bodies
  • Secrets, credentials, or internal tokens
  • Provider-specific payloads that have not been redacted

Privacy flags

  • metadata.captureInput: true or false
  • metadata.captureOutput: true or false
  • metadata.redactionStatus: none, partial, or redacted
  • Use redacted span payloads in demos and shared incidents

Coming later

  • Packaged SDKs and framework auto-instrumentation
  • Slack, email, and PagerDuty delivery
  • Automated regression comparisons
  • Provider-level cost reconciliation