Developer docs

Agent reports

Turn meaningful outcomes from coding agents, monitors, scheduled jobs, and standups into a private audio briefing. AudioReality receives bounded reports; it never operates the systems that sent them.

Start in 30 seconds

Create or reveal a show token on the show's Agents tab in the app: open Shows, choose Edit for the show, then open Agents. The Agents tab appears only when agent reports are enabled (FEATURE_AGENT_REPORTS). If there is no Agents tab for the show, agent keys are not available yet. Agent reports are a deployment setting, not a per-show switch: when the flag is off, the Agents tab is hidden and the v1 endpoints answer 404. Then verify it from your terminal.

export AUDIOREALITY_TOKEN=ar_show_...
npx -y @audioreality/cli doctor

Use the Report v1 contract directly, or choose the integration that matches where your work already runs.

Choose an integration

Public shows

Agent material can feed public programming. A research show takes agent content over public sources, a team standup airs status reports, and a daily digest workflow files content items on a schedule. The pattern stays the same in each case: agents file material, hosts narrate it, and the show owner approves each episode before it reaches listeners.

Nothing airs on its own. Material filed to a public show waits for review, and the reply marks it held for review. Approval is one episode at a time, by the show owner. A group approval does not make an episode public. Read the public-show guide or publish a show.

Status lifecycle

okWork completed cleanly.
warnWork completed with degradation, partial output, or a recovered issue.
failWork did not complete or a required check failed.
needs_decisionProgress is blocked on a human choice or approval.

Give a run a stable run_id. A later status for that same source and run supersedes the earlier report, resolving a pending decision.

Privacy baseline

Only submitted fields leave the machine. Details are disabled by default, content is purged after 14 days, raw text never enters evergreen memory, and AudioReality never trains on it.