Docs/Intelligence

Behaviors

Overview

Behaviors let Auxiora act on your behalf without waiting for a prompt. A behavior is a reusable automation rule that fires on a schedule, polls a condition, triggers at a specific time, or reacts to an event. Every execution is audit-logged and respects your trust settings.

Behavior Types

Type	Trigger	Internal Name	Use Case
Scheduled	Cron expression	`scheduled`	"Every weekday at 9am, summarize my unread emails"
Monitor	Polling interval (60s -- 24h)	`monitor`	"Alert me when Bitcoin drops below $50k"
One-Shot	Fires once at a specific time	`one-shot`	"Remind me to call the dentist at 3pm"
Event	Condition-based trigger on incoming events	`event`	"When a PR is merged to main, notify me on Telegram"

Polling Intervals (Monitor)

Monitors poll at a configurable interval between 60 seconds and 24 hours (86,400 seconds). Intervals shorter than 60 seconds are rejected to prevent excessive resource usage.

Event Triggers

Event behaviors fire when an incoming event matches a set of conditions. Each condition targets a field with an operator (equals, contains, startsWith, endsWith, gt, lt, exists) and a value. Conditions can be combined with and or or logic.

Creating Behaviors

Via Dashboard

Navigate to Behaviors in the sidebar, click Create, select a type, configure the trigger and action prompt, choose a delivery channel, and save.

Via CLI

# Scheduled: daily standup summary (weekdays at 9am)
auxiora behaviors create \
  --type scheduled \
  --cron "0 9 * * 1-5" \
  --prompt "Summarize my calendar and unread messages for today"

# Monitor: PR approval watch (every 5 minutes)
auxiora behaviors create \
  --type monitor \
  --interval 300 \
  --prompt "Check if my open GitHub PRs have been approved"

# One-shot reminder
auxiora behaviors create \
  --type one-shot \
  --at "2026-02-25T15:00:00" \
  --message "Call dentist"

# Event: notify on deployment failure
auxiora behaviors create \
  --type event \
  --source github \
  --event workflow_run \
  --condition "conclusion equals failure" \
  --prompt "Summarize the deployment failure and suggest fixes"

Via API

POST /api/v1/behaviors
Content-Type: application/json

{
  "type": "scheduled",
  "action": "Summarize my calendar and unread messages for today",
  "schedule": {
    "cron": "0 9 * * 1-5",
    "timezone": "America/New_York"
  },
  "channel": {
    "type": "telegram",
    "id": "my-chat-id",
    "overridden": false
  }
}

Managing Behaviors

CLI Commands

Command	Description
`auxiora behaviors list`	Show all behaviors (filter with `--type` or `--status`)
`auxiora behaviors pause <id>`	Pause a behavior (stops executing but retains config)
`auxiora behaviors resume <id>`	Resume a paused behavior
`auxiora behaviors delete <id>`	Permanently remove a behavior

Dashboard

The Behaviors page displays all behaviors in a filterable list. Each row shows the behavior type, status, last run time, run count, and failure count. Actions (pause, resume, delete, run now) are available inline.

Status Transitions

A behavior moves through these statuses:

active -- running on its configured trigger.
paused -- temporarily stopped. Can be resumed.
missed -- a one-shot whose fire time passed while the system was offline.
deleted -- permanently removed.

Error Handling

Auto-Pause on Repeated Failures

If a behavior fails 3 consecutive times (configurable via maxFailures), it is automatically paused to prevent runaway errors. The failure count resets to zero on any successful execution.

Execution Timeout

Each behavior execution has a 60-second timeout. If the action does not complete within this window, it is treated as a failure.

Active Behavior Limit

A maximum of 50 active behaviors can run simultaneously. This limit prevents resource exhaustion on the host machine. Paused and deleted behaviors do not count toward the limit.

Retry on Transient Failure

When a behavior execution fails, it is retried once after a 30-second delay before recording the failure. This catches transient network issues without requiring manual intervention.

Durable Job Queue

Behaviors are backed by a SQLite-based durable job queue that survives process crashes and restarts. This is the same job queue described in the Orchestration documentation.

Crash Recovery

When Auxiora restarts, any jobs that were in running state are automatically reset to pending with an incremented attempt counter. This means a behavior execution that was interrupted by a crash will be retried on the next startup.

Exponential Backoff

Failed jobs are retried with exponential backoff: the delay before the Nth retry is 2^attempt * 1000 milliseconds. After 3 attempts (configurable), the job is moved to the dead-letter state.

Non-Retryable Errors

Some errors are marked as non-retryable (e.g., invalid configuration, missing credentials). These skip the retry queue and are recorded as dead immediately.

Checkpoint Support

Long-running behavior handlers can call ctx.checkpoint(data) to persist intermediate state. If the process crashes, the handler resumes from the last checkpoint via ctx.getCheckpoint() rather than starting over.

Fallback Mode

If the SQLite job queue is unavailable (e.g., during first run before initialization), behaviors fall back to an in-memory execution queue. This queue does not survive crashes but ensures behaviors still function.

Use Cases

1. Morning Briefing

A scheduled behavior at 8:00 AM pulls your calendar events, email summaries, weather forecast, and top news stories. The compiled briefing is delivered via Telegram before you start your day.

auxiora behaviors create \
  --type scheduled \
  --cron "0 8 * * *" \
  --prompt "Generate my morning briefing: calendar, unread emails, weather, top news" \
  --channel telegram

2. Competitor Monitoring

A monitor behavior checks competitor websites every 6 hours, diffs the content against the previous snapshot, and summarizes any changes.

auxiora behaviors create \
  --type monitor \
  --interval 21600 \
  --prompt "Check competitor.com/pricing for changes and summarize any differences" \
  --channel slack

3. Meeting Prep

A one-shot reminder fires 30 minutes before each meeting. It pulls attendee information from Google Workspace, related Notion documents, and recent email threads to prepare a context brief.

auxiora behaviors create \
  --type one-shot \
  --at "2026-02-25T13:30:00" \
  --prompt "Prep me for my 2pm meeting: pull attendee info, Notion project notes, recent email threads"

4. Health Check

A monitor behavior pings your production API health endpoint every 5 minutes and alerts you on Telegram if it returns a non-200 status.

auxiora behaviors create \
  --type monitor \
  --interval 300 \
  --prompt "Check https://api.example.com/health — alert me if status is not 200" \
  --channel telegram