Getting Started

Quickstart

Build and run your first AI-powered automation workflow in minutes — fully local, privacy-first, and under your control.

This platform is designed around local AI agents and deterministic workflows. Nothing is hidden, nothing is sent to third-party SaaS platforms by default.

You define agents, compose workflows, and execute tasks using a runner that you own and operate.

1. Create an Agent

An agent represents an autonomous unit that can reason, use tools, and execute steps. Agents can send emails, read/write files, browse the web, or call language models.

{
  "name": "Research Agent",
  "description": "Summarizes and analyzes information",
  "tools": ["llm", "email", "file", "browser"]
}

2. Define a Workflow

A workflow is a sequence of steps executed in order. Each step produces structured output that feeds into the next step.

{
  "name": "Daily Report Workflow",
  "steps": [
    { "type": "llm", "prompt": "Summarize today's updates" },
    { "type": "file", "action": "write", "path": "./runtime/report.txt" },
    { "type": "email", "to": "me@example.com" }
  ]
}

3. Execute with the Runner

The runner picks up tasks, executes each step, logs output, and updates execution state in real time.

npm run worker
Task claimed: workflow_01
→ Executing step: LLM prompt
→ Writing output to runtime/

What Happens When You Run a Workflow

Every workflow execution follows a deterministic lifecycle. Understanding this flow helps you debug failures, optimize steps, and reason about agent behavior with confidence.

When you click “Run Workflow”, the platform follows a deterministic execution path. Here’s what happens under the hood:

Workflow Initialization

Runner initializes the workflow and its steps.

Task Claiming

Runner claims the task for execution.

Step Runner Initialization

Step runner sets up execution context.

Tool Execution

Each step is dispatched to the appropriate tool implementation.

Logs

Execution logs are recorded for each step.

Status

Task status is updated in real-time.

How the System Works (Mental Model)

Think of the platform as three coordinated systems working together:

  • Workflows define what should happen
  • Tasks are runtime executions of workflows
  • Agents & Tools perform each step deterministically

🎯 That’s it. You now have a fully local AI automation system running under your control. No vendor lock-in, no black boxes, no hidden costs.