Skip to main content
Autonomous Agent Graphs run without a live user. External systems, schedules, or events start a session; the graph executes tools, routing, and optional registry agent calls in the background.

Triggers overview

API, cron, and background-task trigger types.

Deploy a trigger

Wire a build to API or Cron ingress.

Graph Studio

Design automation logic on the canvas.

Observability

Monitor runs, logs, and token usage.

​
What Autonomous graphs do

Workspace Integrations hub with Triggers tab

Integrations — Triggers tab for Autonomous deploy targets

​
Trigger types

See Triggers overview for the full automation hub.

​
Design workflow

  1. Create an Autonomous Agent Graph from Workspace Home.
  2. In Graph Studio, model the automation as nodes — typically Start → agent/tool chain → End.
  3. Attach tools for each external action (CRM update, email send, data fetch).
  4. Use variables to pass trigger payload fields into tool inputs.
  5. Click Save, then Build (Builds overview).
  6. Assign the build to DEV / UAT / PROD.
  7. Click Deploy and choose a target:

​
Deploy an API trigger

  1. Build and assign your Autonomous graph to the target environment.
  2. Click Deploy → Deploy as API.
  3. Select build version and confirm trigger configuration.
  4. Copy the webhook URL and authenticate with your workspace API key.
  5. POST a JSON payload from your upstream system to start a run.
See Deploy a trigger and Trigger APIs.

​
Deploy a cron schedule

  1. Build and assign the graph.
  2. Click Deploy → Cron job.
  3. Enter the cron expression and timezone.
  4. Select environment and build version.
  5. Confirm — Phinite runs the graph on schedule.
See Trigger APIs — Cron.

​
Best practices

  • Design actions to be idempotent — triggers may retry or duplicate.
  • Return structured outputs and capture variables for downstream auditing.
  • Set timeouts and error branches so failed tool calls do not leave silent failures.
  • Test with sample payloads in DEV before promoting builds to PROD.
  • Monitor runs in Observability — there is no user to report errors.
Autonomous graphs run with production credentials in PROD. Validate env variables and integration auth in UAT first.