Getting Started

Install QUESTPIE Autopilot, create your first company, and run AI agents in under 60 seconds. Zero infrastructure required.

Install Autopilot, scaffold a company, and give your first intent. One Bun process. One SQLite file. Zero infrastructure.

Prerequisites

Bun v1.3+ -- runtime, package manager, and test runner
terminal
```
curl -fsSL https://bun.sh/install | bash
```
Authentication (choose one):
- autopilot provider set openrouter --api-key sk-or-... -- use your OpenRouter API key (recommended)
- OPENROUTER_API_KEY -- API key from openrouter.ai/keys (alternative)
Git -- required for versioning company state and code operations

That's it. No Docker. No Postgres. No Redis. No vector DB.

Quick Start

terminal

bun add -g @questpie/autopilot
autopilot init my-company
cd my-company

# Authenticate (choose one)
autopilot provider set openrouter --api-key sk-or-...    # Set your OpenRouter API key
# OR
export OPENROUTER_API_KEY=sk-or-...  # Use API key

# Start the orchestrator + dashboard
autopilot start

# Open dashboard
open http://localhost:3000

# Send your first task
autopilot ask "Build me a landing page"

# Watch agents work in real-time
autopilot attach max

What Just Happened

autopilot init scaffolded a complete company filesystem from the solo-dev template -- agents, workflows, knowledge, skills, and a living dashboard.
autopilot start launched the orchestrator: filesystem watcher, workflow engine, scheduler, webhook server, REST API, and SSE stream. One process.
autopilot ask gave your intent to the CEO agent, who decomposed it into scoped tasks and assigned them to the right agents via workflow rules.
autopilot attach connected you to Max's live session stream -- like kubectl logs -f for AI agents. Ctrl+C to detach; the agent keeps working.

Project Structure

After autopilot init, you get a complete company filesystem:

my-company/

company.yaml                  # Company configuration
team/
  agents.yaml                 # AI agent definitions (8 default agents)
  humans.yaml                 # Human team members
  roles.yaml                  # RBAC role definitions
  schedules.yaml              # Cron-triggered agent jobs
  webhooks.yaml               # External webhook integrations
  workflows/
    development.yaml          # 12-step dev workflow
    marketing.yaml            # 7-step marketing workflow
    incident.yaml             # 8-step incident response
  policies/
    approval-gates.yaml       # Human approval requirements
knowledge/                    # Searchable knowledge base (markdown)
  brand/                      # Brand guidelines
  business/                   # Business context
  technical/                  # Architecture & conventions
  onboarding/                 # Agent onboarding docs
  integrations/               # Integration guides
skills/                       # Agent skills (agentskills.io format, 20 built-in)
dashboard/                    # Living dashboard configuration
  groups.yaml                 # Pin groups
  pins/                       # Agent-created dashboard pins
  widgets/                    # Custom React widgets
  pages/                      # Custom dashboard pages

The above directories are created by the template. The following runtime directories are created automatically by autopilot init and autopilot start:

Runtime directories

logs/                         # Activity & session logs (runtime)
context/                      # Agent memory & indexes (runtime)
  memory/{agent}/             # Per-agent persistent memory
secrets/                      # Encrypted secrets (runtime)
projects/                     # Project workspaces (runtime)
artifacts/                    # Generated apps & content (runtime)
.data/
  autopilot.db                # SQLite database (FTS5, embeddings, auth)

Tasks, messages, and activity are stored in SQLite (.data/autopilot.db), not as YAML files. The .data/ directory is auto-created on first autopilot start. If deleted, it is rebuilt from the configuration files.

Configuration

The root company.yaml controls company identity, agent runtime settings, and budget limits:

company.yaml

name: "My Company"
slug: "my-company"
description: "A solo developer shop powered by QUESTPIE Autopilot"
timezone: "UTC"
language: "en"
languages: ["en"]

owner:
  name: "Founder"
  email: "founder@example.com"
  notification_channels: []

settings:
  auto_assign: true
  require_approval: ["merge", "deploy", "spend", "publish"]
  max_concurrent_agents: 4
  agent_provider: "tanstack-ai"
  agent_model: "claude-sonnet-4-6"
  budget:
    daily_token_limit: 2000000
    alert_at: 80

integrations: {}

Agents are defined in team/agents.yaml. Each agent has an ID, role template, filesystem scope, and available tools:

team/agents.yaml (excerpt)

agents:
  - id: ceo
    name: "CEO Agent"
    role: meta
    description: "Decomposes high-level intents into tasks, manages company structure"
    fs_scope:
      read: ["/**"]
      write: ["/team/**", "/dashboard/**"]
    tools: ["task", "message", "pin", "search_index", "web_search"]

  - id: max
    name: "Max"
    role: developer
    description: "Implements features, writes code, creates branches and PRs"
    fs_scope:
      read: ["/knowledge/technical/**", "/projects/**"]
      write: ["/projects/*/code/**"]
    tools: ["task", "message", "pin", "search_index", "fetch", "web_search"]

Start with 2-3 agents, add as you need. You choose the names, the roles, and the tools.

First Commands

Start the orchestrator

terminal

cd my-company

# Authenticate (choose one)
autopilot provider set openrouter --api-key sk-or-...    # Set your OpenRouter API key
# OR
export OPENROUTER_API_KEY=sk-or-...  # Use API key

autopilot start

This starts the filesystem watcher, scheduler, webhook server (port 7777), REST API + SSE stream (port 7778), and initializes the SQLite database. It runs until you stop it.

Open the dashboard

terminal

autopilot dashboard

# Or open directly
open http://localhost:3000

The Living Dashboard shows agent activity, task status, board pins, and approval gates. It updates in realtime via SSE. Agents can create custom widgets and pages that appear automatically.

Give your first intent

terminal

autopilot ask "Set up a Next.js project with authentication"

# The CEO agent decomposes this into scoped tasks:
# 1. Sam (strategist) scopes the feature and writes a spec
# 2. Alex (planner) creates a file-level implementation plan
# 3. Max (developer) writes the code and opens a PR
# 4. Riley (reviewer) reviews for quality and spec compliance
# 5. You approve the merge (human gate)
# 6. Ops (devops) deploys to staging, then production

Watch agents work

terminal

# List all agents and their current status
autopilot agents

# Live-stream an agent's session (like kubectl logs -f)
autopilot attach max

# Compact mode -- one line per tool call
autopilot attach max --compact

# Only show tool calls, skip thinking
autopilot attach max --tools-only

Handle approvals

terminal

# Check your inbox for items needing attention
autopilot inbox

# Approve a task at a human gate
autopilot approve TASK-001

# Reject with feedback -- agent will rework
autopilot reject TASK-001 "Needs error handling for edge cases"

Next Steps

Tools -- the 12 unified tools agents use (the core differentiator)
Living Dashboard -- custom widgets, theme overrides, agent-created pages
Architecture -- six-layer stack, zero-infra design, orchestrator internals
Workflows -- how tasks flow through development, marketing, and incident pipelines
Skills -- teach agents new capabilities via SKILL.md files
CLI Reference -- all available commands and options

Getting Started

On this page