Human Control

AI Agents That Know When to Ask for Help

Autonomous doesn't mean unsupervised. Orchestratia gives you plan approval, intervention queues, and mobile control via Telegram.

Fully Autonomous Is Rarely the Goal

AI coding agents are powerful but imperfect. They occasionally:

Make architectural decisions you disagree with
Get stuck on ambiguous requirements
Need access to resources they can't reach
Produce code that technically works but doesn't match your intent

The answer isn't to remove autonomy — it's to add structured checkpoints where humans can guide, approve, and correct.

Structured Human Intervention

Orchestratia provides three levels of human control:

1. Plan Mode

For complex tasks, agents submit a plan before writing code. You review and approve the approach — or send it back with revisions.

2. Help Requests

When an agent hits a roadblock, it transitions to needs_human state and creates an intervention. The question appears in your dashboard queue and as a Telegram notification.

3. Task Notes

You can push notes to running agents at any time — guidance, context, corrections. Notes are delivered in real-time via WebSocket and displayed in the agent's session.

graph TB
    Agent["Agent Working"] -->|"Stuck"| Help["Request Help"]
    Agent -->|"Complex task"| Plan["Submit Plan"]
    Admin["Admin"] -->|"Send note"| Agent
    Help --> Queue["Intervention Queue"]
    Plan --> Review["Plan Review"]
    Queue -->|"Respond"| Agent
    Review -->|"Approve"| Agent
    Review -->|"Revise"| Plan

    style Agent fill:#FAF8F5,stroke:#E8E3DA
    style Help fill:#F0F9F7,stroke:#2A9D88,stroke-width:2px
    style Plan fill:#F0F9F7,stroke:#2A9D88,stroke-width:2px
    style Queue fill:#FAF8F5,stroke:#E8E3DA
    style Review fill:#FAF8F5,stroke:#E8E3DA
    style Admin fill:#F0F9F7,stroke:#2A9D88,stroke-width:2px

Telegram Integration

Every intervention, plan submission, and task completion can trigger a Telegram notification. You can:

View live agent output in Telegram
Tap buttons to approve or deny permissions
Respond to help requests from your phone
Focus on specific sessions for detailed monitoring

Plan Mode

Agents submit plans before coding. You approve, revise, or reject the approach.

Intervention Queue

When agents need help, questions queue up in the dashboard for your response.

Real-Time Notes

Push guidance to running agents at any time. Notes appear instantly in their session.

Telegram Control

View output, approve permissions, and respond to agents from your phone.

Stay in control of your AI agents

Set up plan mode, intervention workflows, and Telegram notifications.

Get Started View Documentation