Scheduling

HiveMind OS runs as a background daemon, which means it can execute tasks on your behalf even when the UI is closed. The scheduler supports several trigger types so you can automate anything from daily reminders to event-driven workflows.

Task Schedule Types

Tasks use the TaskSchedule enum with three variants:

Type	Description	Example
Once	Run immediately, one time only	Fire-and-forget tasks
Scheduled	Run once at a specific timestamp	run_at_ms: 1733076000000 — run at a specific epoch time
Cron	Recurring cron expression	expression: "0 9 * * MON-FRI" — every weekday at 9 AM

Workflow Trigger Types

Workflows support a broader set of triggers:

Trigger	Description
manual	Started by user action
incoming_message	Reacts to messages from connectors (email, Slack, etc.)
event_pattern	Matches patterns from the internal event bus
mcp_notification	Triggered by MCP server notifications
schedule	Recurring via cron expression

Creating a Scheduled Task in YAML

Add a task definition to your config.yaml or create a standalone task file:

tasks:
  - name: morning-pr-review
    schedule:
      cron:
        expression: "0 9 * * MON-FRI"
    agent_config:
      strategy: react
      model_role: primary
      tools: [mcp:github]
    input:
      prompt: "Check open PRs in my repos and summarize anything needing review."
    data_class: INTERNAL
    retries:
      max: 2
      backoff: 5m
    timeout: 10m

Managing Tasks

From the UI: Open the Tasks view in the sidebar to see all scheduled, running, and completed tasks. Click any task to view its logs, edit its schedule, or pause it.

You can also manage scheduled tasks from the Workflows page in the UI.

Event-Driven Workflows

Workflows can react to events using the incoming_message, event_pattern, and mcp_notification trigger types:

# workflow.yaml
name: auto-triage-issues
trigger:
  type: mcp_notification
  server: github
  method: "issues.opened"
steps:
  - name: triage
    agent_config:
      strategy: react
      tools: [mcp:github]
    input:
      prompt: "Triage this new issue: label it, estimate priority, and notify me if urgent."

The internal event bus connects MCP notifications, task completions, and user actions — so you can chain tasks together.

Task Persistence & Recovery

Tasks are durable. The scheduler persists task state to disk, so tasks survive daemon restarts, OS reboots, and sleep/wake cycles. On startup, the scheduler detects any tasks left in a running state (stale from a previous crash) and automatically resets them to pending so they can be retried.

Resource governance

The scheduler enforces concurrency limits and per-provider rate limits. Tasks also inherit their creator's data-classification context — a task created in a CONFIDENTIAL session won't leak data to public channels.