ancha traces

ancha traces lists the traces in your workspace, newest run first. ancha traces show previews the test one trace makes: what the agent is told, the world, the checks with what production did, and the questions one trace cannot settle. Nothing is saved until you run ancha pull.

ancha traces [options]
ancha traces show <id>
ancha traces diff <trace file> [attempt]

Option	Description
`--failed`	Only traces where production failed a check
`--passed`	Only traces where production passed every check
`--agent <name>`	Only this agent’s traces
`--tool <tool>`	Only traces that called this tool, such as `slack`
`--env <name>`	Only traces from this deployment environment: OpenTelemetry’s `deployment.environment`, or Langfuse’s environment
`--limit <n>`	At most this many (default: 20)

ancha traces show takes a trace’s id, its first 8 characters, or the trace id your tracing tool shows. Sign in with ancha login first, or set ANCHA_SERVER_URL and ANCHA_API_KEY. With --json, both print the workspace’s data as one JSON object.

Example

$ ancha traces --failed
2 traces, newest first:
  4735f9ca  2026-10-21 13:59 UTC  acme-oncall  datadog, github, slack, pagerduty  production failed 3 checks  test saved
  764cfbcc  2026-10-21 13:59 UTC  acme-oncall  datadog, github, slack, pagerduty  production failed 2 checks  test saved

$ ancha traces show 4735f9ca
trace 4735f9ca (85ff075cdaec081c188c079b0eab6abd): acme-oncall, started 2026-10-21 13:59 UTC, 74 records
what arrived:
  datadog: 14 API calls, rebuilt
  github: 9 API calls, rebuilt
  slack: 7 API calls, rebuilt
  pagerduty: 6 API calls, rebuilt
  17 model calls, with what the agent was told
the test it makes:
  the agent is told: Orders requests are failing after a deploy. …
  the world:
    Slack: #orders-incident, 1 person, 1 earlier message
    GitHub: acme/orders, 2 commits on main
    At 91s, someone posts "Status request: what is the current customer impact, …"
    A merge to the default branch deploys production in 15s
    The problem is on from the start; it ends when production has the fix in config/runtime.ts, not before 128s, then recovers over 20s
  checks:
    answered-msg-status-request     failed in production  (the agent answers "Status request: …" in its thread)
    problem-resolved                failed in production  (the problem ends)
  questions one trace cannot settle (the test runs with each default until you answer):
    deploy-repo_orders: What happens when a pull request merges to main in acme/orders?
      default: It deploys production in about 15 seconds
      answer: ancha pull 4735f9ca --answer deploy-repo_orders=<seconds until production has the merge>, or deploy-repo_orders=none
not saved yet: ancha pull 4735f9ca saves it as a test and writes it here

A tool marked not rebuilt yet is left out of the test. A note under what arrived says what a trace is missing, such as response bodies or what the agent was told, and how to send it. See From a production trace to a failing test.

Compare an attempt with production

When a test doesn’t fail the way production did, compare the calls. ancha traces diff reads a trace file and one attempt of a finished run on this computer:

ancha traces diff trace.otlp.jsonl latest/trial-tom-replies-1

Output

production trace.otlp.jsonl: 14 calls; attempt run_20261107-052607_e3cc66e6ab5307b4/trial-tom-replies-1: 14 calls
only in production:
  gmail GET /gmail/v1/users/me/threads/:id
  gmail POST /gmail/v1/users/me/drafts
only in the attempt:
  gmail GET /gmail/v1/users/me/history ×2
the order parts at call 11: production gmail GET /gmail/v1/users/me/threads; the attempt gmail POST /gmail/v1/users/me/threads/:id/modify

Calls compare by tool, method and path, with the record IDs in a path written :id. The attempt is <run id or latest>/<trial directory>; a run alone means its first attempt. With --json, the result has production and attempt (each with its call count), onlyInProduction and onlyInAttempt (each call with its count) and firstDifference (index, production and attempt), which is null when the orders match. It exits 0 either way. A call production made and the attempt didn’t, such as labels created at setup or a settings read, usually points at what the seed lacks. For the agent’s own tools, ancha explain <run> --trial <n> lists every answer that was not production’s.

Get started

Guides

Concepts

Tools

CLI

Reference

Cloud

Help

ancha traces

Example

Compare an attempt with production

​Example

​Compare an attempt with production

Example

Compare an attempt with production