Configure Ava Autonomous Capabilities

This guide covers three autonomous capabilities in Ava: self-scheduling recurring tasks, storing and recalling information with persistent memory, and gateway health monitoring with auto-remediation. After reading it, you will be able to configure all three for your project.

Prerequisites

• protoLabs Studio running with a project open
• Ava chat accessible (/chat route)
• .automaker/ava-config.json writable for the project (created automatically on first save)

Set Up Self-Scheduling

Ava can schedule recurring tasks that invoke a stored prompt on a cron schedule or fixed interval. Tasks persist across server restarts via .automaker/ava-tasks.json.

Enable the scheduling tool group

The scheduling group is enabled by default. To verify or explicitly enable it:

{
  "toolGroups": {
    "scheduling": true
  }
}

Save this to .automaker/ava-config.json in your project root.

Schedule a task via chat

In the Ava chat, ask Ava to schedule a task:

Schedule a daily board summary every weekday morning at 9am.
Prompt: "Summarize the current board state: how many features are in progress,
how many are blocked, and any features that have been stuck for more than 2 days."

Ava calls schedule_task with a cron expression and stores the task. The task ID is prefixed ava: automatically.

To schedule on a fixed interval instead of a cron:

Check for stuck features every 2 hours.
Prompt: "List any features that have been in_progress for more than 6 hours without an update."

View scheduled tasks

List my scheduled tasks.

Ava calls list_scheduled_tasks and returns task IDs, schedules, last/next run times, and failure counts.

Cancel a task

Cancel the task ava:daily-board-summary.

Ava calls cancel_task. Only ava:-prefixed tasks can be cancelled.

Trigger a task immediately

Run the ava:daily-board-summary task now.

Ava calls trigger_task and returns the result of the prompt invocation.

Cron expression reference

Expression	Meaning
0 9 * * 1-5	Weekdays at 9:00 AM
0 */2 * * *	Every 2 hours
*/30 * * * *	Every 30 minutes
0 0 * * *	Daily at midnight

Use Persistent Memory

The memory tool group lets Ava store and retrieve information across chat sessions. Memory is stored at .automaker/ava-memory.json.

Enable the memory tool group

The memory group is enabled by default. To verify or explicitly enable it:

{
  "toolGroups": {
    "memory": true
  }
}

Store information

Ask Ava to remember something:

Remember that our deploy process requires a manual approval step in the #deployments channel.
Key: deploy-process
Tags: ops, deploy

Ava calls remember. If the key already exists, the content and tags are updated.

Recall information

What do you know about our deploy process?

Ava calls recall with the query. Results are returned in ranked tiers:

1. Exact key match
2. Tag match
3. Substring match in key or content

Remove a memory

Forget the deploy-process memory — it's outdated.

Ava calls forget with the exact key.

Common memory patterns

Store preferences and decisions that Ava should carry across sessions:

Remember that we use conventional commits (feat/fix/refactor/chore) and squash-merge feature PRs.
Key: commit-conventions
Tags: git, workflow

Remember that the staging environment URL is https://staging.example.com and deploys happen automatically on push to staging.
Key: staging-env
Tags: ops, deploy, staging

Enable Gateway Auto-Remediation

The Ava Gateway runs periodic heartbeat checks on the board. By default it only alerts to Discord. With the gatewayAutoRemediate feature flag enabled, the gateway takes corrective action automatically.

Enable the feature flag

In the UI: Settings > Developer > Feature Flags > Gateway Auto-Remediate.

Or via API:

{
  "featureFlags": {
    "gatewayAutoRemediate": true
  }
}

When disabled (default), the gateway posts alerts to Discord only. When enabled, the GatewayActionExecutor runs after each heartbeat and can take up to 3 actions per cycle.

Available auto-remediation actions

Action	What it does	When it triggers
unblock_feature	Resets blocked features with retryable errors back to backlog	Feature blocked > 30 min with retryable classification
retry_agent	Restarts a failed agent session	Feature failed with transient error (rate limit, timeout)
merge_ready_pr	Merges PRs where all checks pass	PR in review with all CI green, no unresolved threads

Action budget

The executor enforces a budget of 3 actions per heartbeat cycle. Actions beyond the budget are logged but not executed. This prevents runaway remediation loops.

All executed actions are recorded in an append-only audit array on the gateway status object and emitted as ava-gateway:action-executed events.

Configure the infra channel

Set the DISCORD_CHANNEL_INFRA environment variable to your Discord infra channel ID:

DISCORD_CHANNEL_INFRA=1469109809939742814

The gateway posts startup messages, heartbeat alerts, and action summaries to this channel.

View gateway status

GET /api/ava/status

Returns the current gateway state:

{
  "initialized": true,
  "listening": true,
  "lastHeartbeat": "2026-03-21T09:00:00.000Z",
  "lastHeartbeatStatus": "ok",
  "totalHeartbeats": 42,
  "totalAlerts": 1,
  "circuitBreaker": {
    "isOpen": false,
    "failureCount": 0
  }
}

Circuit breaker behavior

The gateway includes a circuit breaker that opens after 5 consecutive failures. When open, heartbeat evaluations and auto-remediation are skipped for 5 minutes. The circuit resets automatically after the cooldown.

If alerts are firing repeatedly, check /api/ava/status for circuitBreaker.isOpen and failureCount.

Next steps

• Timer Registry — View and control scheduled tasks from the Ops Dashboard (internal docs: docs/internal/server/timer-registry.md)
• Ava Chat System — Architecture reference for all Ava tool groups (internal docs: docs/internal/server/ava-chat.md)