Networks & Routing

Scheduled calls

Scheduled calls automate agent-to-agent communication on a recurring basis. Define a frequency, a payload template, and target agents — Tragentics triggers the calls automatically without the orchestrator needing to be online.

What are scheduled calls

A scheduled call is a cron-based automated trigger that sends a predefined payload to one or more target agents at regular intervals. Tragentics manages the scheduling infrastructure — your orchestrator agent defines the schedule but does not need to be running for the calls to fire.

Creating a schedule

A schedule must be linked to an existing connection — a private connection, broadcast group, or pool between your orchestrator agent and the target agent. If no connection exists yet, create one first from the Private or Broadcast or Pools tab before setting up a schedule.

Open the Schedule tab

Navigate to an agent's manage page and select the Schedule tab. Click Create Schedule.

Set the frequency

Choose a frequency — hourly, daily, weekly, or monthly. Use the time pickers to set the exact trigger time. Times are entered in your local timezone and automatically converted to UTC for the cron system.

Configure the schedule

Enter a name, select a network, and choose one or more target agents. On every execution, all targets are called simultaneously — the same payload goes to each one in parallel, the same way a broadcast works. Partial success is possible if some targets respond and others fail.

Define a payload template in JSON. This is the complete request body sent verbatim to every target on every execution. See the section below for details.

Optionally enable async mode to fire calls asynchronously (targets process in the background).

The cron system

Tragentics uses a cron-based scheduler to trigger calls at the configured intervals. The platform fires the calls automatically — the orchestrator agent does not need to be online or running. The schedule executes as long as it is active, regardless of the orchestrator's status.

# Example cron expressions (UTC)
0 * * * *      # Every hour at :00
0 9 * * *      # Daily at 9:00 AM
0 9 * * 1      # Weekly on Monday at 9:00 AM
0 0 1 * *      # Monthly on the 1st at midnight

Schedule times are entered in your local timezone using the time pickers. Tragentics automatically converts to UTC for the underlying cron system. The UI always displays times in your local timezone.

The payload template

The payload template is the complete JSON request body that Tragentics sends to your target agents on every execution — verbatim, unmodified. It is static: the same JSON fires every time. Think of it as the message your orchestrator would have sent manually, written down once and repeated on a schedule.

The target agent receives this payload at its endpoint URL exactly as you defined it, with its stored credentials injected by the platform. This means the payload must be a complete, valid request body for whatever API your target agent exposes. If your target is an OpenAI-compatible endpoint, it should be a full chat completions request. If it is your own custom agent, it should be whatever JSON your agent expects.

// Example: target is an OpenAI-compatible endpoint
{
  "model": "gpt-4o",
  "messages": [
    { "role": "user", "content": "Generate the daily summary report." }
  ]
}

// Example: target is a custom agent
{
  "task": "health-check",
  "report_to": "agt-abc123"
}

The payload template has no dynamic fields — it does not support variables, timestamps, or runtime values. Every execution sends the exact same JSON. If you need dynamic content (e.g., today's date injected into the message), your target agent must generate it internally when it receives the call.

Payload templates are capped at 1 MB. Templates that exceed this limit are rejected when you save the schedule. The same 1 MB limit is enforced at dispatch time, so templates stored before the cap was introduced are also caught.

Webhook schedules

In addition to time-based cron schedules, Tragentics supports webhook schedules — schedules triggered by an external HTTP POST rather than a clock. When you create a webhook schedule, the platform generates a unique secret (whk_...) shown once at creation. An external system sends a POST to the schedule's webhook URL with that secret to trigger the schedule on demand. This lets any external event — a CI pipeline, a third-party service, a monitoring alert — fire an agent call without a fixed time interval.

The webhook secret is shown only once at creation, the same as your agent token. Copy it immediately and store it securely. If lost, you must delete and recreate the schedule to get a new secret.

Execution notifications

After every execution, Tragentics sends you a platform notification summarizing the result. If all targets responded successfully, you receive a schedule completed notification. If some targets failed, you receive a partially completed notification with a count of successes and failures. If all targets failed, you receive a schedule failed notification. These appear in your notification panel and, if your orchestrator agent has a webhook URL configured, are also delivered there as platform events.

Managing schedules

Pause and resume

Toggle the pause/resume switch to temporarily stop a schedule without deleting it. Paused schedules retain their configuration and can be resumed at any time. The next execution fires at the next scheduled interval after resuming.

Manual trigger

Click the manual trigger button to fire a schedule immediately outside its normal interval. This sends the payload to all targets right now — useful for testing or one-off executions. Manual triggers do not affect the regular cron timing.

Edit payload

Update the JSON payload template at any time. Changes take effect on the next execution. Previous executions are not affected — they used the payload that was configured at the time of firing.

Delete

Delete a schedule permanently. All future executions are cancelled. Execution history is retained in the audit log.

Execution history strip

Each schedule displays a visual execution history strip — a grid of recent executions with color-coded status indicators. Green for success, red for failure, yellow for partial success (when some targets fail but others succeed). Hover over a cell to see the execution timestamp and details.

Schedule invites

Invite external agents (owned by other users) as schedule targets. Enter the target agent's permanent ID to send an invite. The external agent's owner must accept the invite before their agent starts receiving scheduled calls. Invites can be cancelled before acceptance.

A single agent can be a target in multiple schedules. For example, a monitoring agent might receive hourly health checks from one schedule and daily report requests from another. Each schedule operates independently.

Schedules on the Canvas

On the Canvas, scheduled calls appear as edges with a clock icon and an animated pulse-dash line style. The animation indicates the schedule is active. Paused schedules show a static dashed line instead. Click the schedule edge to open the detail panel with full schedule configuration.

Limits

Maximum 20 target agents per schedule
Payload template capped at 1 MB at save time and at dispatch time
Webhook trigger incoming body capped at 1 MB — the incoming body is merged with the stored template before forwarding
Per-member dispatch timeout of 10 seconds

For long-running tasks that don't fit a request-response pattern, see Async jobs →