Agent Deployments
Deploy agents to production with durable execution, webhook and schedule triggers, quota management, and lifecycle controls.
Agent Deployments turn your configured agent instances into always-on, production-ready services. A deployed agent runs under a durable supervisor that manages execution, enforces rate limits, and responds to triggers automatically.
Deployment Architecture
When you deploy an agent, Fabric creates a durable supervisor that manages the agent's lifecycle:
Deploying an Agent
Configure an Agent Instance
Before deploying, you need a configured agent template instance. Set up the agent's tools, prompts, and behavior in Agent Templates.
Set Deployment Configuration
Configure execution limits for the deployment:
| Setting | Description | Default |
|---|---|---|
| Max Concurrent Executions | How many executions can run at once | 5 |
| Rate Limit Per Minute | Maximum executions per minute | 60 |
| Rate Limit Per Hour | Maximum executions per hour | 500 |
| Daily Execution Limit | Optional daily cap | No limit |
| Monthly Execution Limit | Optional monthly cap | No limit |
Deploy
Click Deploy to start the deployment. Fabric will:
- Create a deployment record
- Start the durable supervisor process
- Activate any triggers defined on the agent instance (webhooks, schedules)
- Return the deployment ID and trigger details
Deployment Statuses
| Status | Description |
|---|---|
| Active | Supervisor running, accepting execution requests |
| Paused | Supervisor paused, no new executions accepted, existing executions may complete |
| Terminated | Supervisor stopped, deployment ended |
Lifecycle Management
Pause
Pausing a deployment:
- Signals the supervisor to stop accepting new executions
- Deactivates all webhook triggers (incoming webhooks are rejected)
- Pauses all schedule triggers
- Running executions may continue to completion
Resume
Resuming a paused deployment:
- Signals the supervisor to accept executions again
- Reactivates webhook triggers that were paused
- Resumes schedule triggers
- The deployment returns to Active status
Terminate
Terminating a deployment:
- Signals the supervisor to shut down
- Deletes all associated schedules
- Removes all trigger records
- Releases the deployment quota
- This action is permanent
Triggers
Triggers define how a deployed agent is automatically invoked. Each deployment can have multiple triggers.
Webhook Triggers
Webhook triggers expose an HTTP endpoint that accepts POST requests:
POST /api/webhooks/agent/{unique-path}| Feature | Description |
|---|---|
| Authentication | HMAC-SHA256 signature verification with a generated secret |
| Replay Protection | Timestamp validation (5-minute window) |
| IP Allowlisting | Optional restriction to specific IP addresses |
| Payload Size Limit | Configurable, up to 10 MB |
| Input Template | Default input that merges with webhook payload |
| Priority | Execution priority (Low, Normal, High, Critical) |
The webhook secret is shown only at creation time. You can regenerate it if needed, which invalidates the previous secret.
Schedule Triggers
Schedule triggers use cron expressions to run agents on a recurring basis:
| Feature | Description |
|---|---|
| Cron Expression | Standard 5-part cron format (minute, hour, day, month, weekday) |
| Timezone | IANA timezone for schedule evaluation (default: UTC) |
| Overlap Policy | If the previous execution is still running, the scheduled run is skipped |
| Catchup Window | Missed schedules within 1 minute are caught up |
Schedule triggers are backed by durable execution infrastructure, providing reliable and fault-tolerant scheduling.
Manual Execution
You can trigger an execution manually at any time through the UI or API:
- Provide optional input data as a JSON object
- Set execution priority (Low, Normal, High, Critical)
- Executions are queued and processed by the supervisor
Execution Management
Execution Flow
- A trigger (webhook, schedule, or manual) creates an execution record
- The execution is queued with the supervisor
- The supervisor enforces rate limits and concurrency controls
- The agent runs with the provided input
- Results are stored in the execution record
Execution Statuses
| Status | Description |
|---|---|
| Pending | Queued, waiting for the supervisor to process |
| Running | Agent execution in progress |
| Completed | Execution finished successfully |
| Failed | Execution encountered an error |
| Cancelled | Execution was cancelled by the user |
Cancelling Executions
You can cancel a pending or running execution from the deployment detail page or via the API.
Quota Management
Deployment quotas prevent resource overuse:
- Deployment Quota — Limits how many active deployments a user or organization can have
- Execution Quota — Per-deployment limits on daily and monthly executions
- Rate Limits — Per-minute and per-hour execution caps
Check quota usage from the deployment detail page or via the API.
Multi-Tenant Support
Agent deployments work in both personal and organization contexts:
- Personal (
/app/agent-deployments) — Your private deployments - Organization (
/app/{org}/agent-deployments) — Organization-scoped deployments visible to members - Deployment quotas are tracked separately per context
- Organization members can manage deployments based on their org role
- Webhook and schedule triggers are tenant-isolated
- Task queue sharding separates personal and organization workloads