QUESTPIE Autopilot

QUESTPIE Autopilot is an AI-native company operating system where AI agents run your company through unified tools, human approval gates, and a self-evolving dashboard. You define agents in YAML, give high-level intents, and approve at explicit gates. Company configuration lives in files you can ls, grep, and version with Git, while runtime data (tasks, messages, activity) is stored in SQLite for fast indexed queries. It runs as a single Bun process with zero infrastructure.

What Makes Autopilot Different

Tools, Not Chat

Agents don't generate text for you to read. They call 12 unified tools -- task(), message(), pin(), search_index(), fetch(), web_search(). Agent thinking is private. Only tool calls produce visible effects. This is not a chatbot wrapper. It's a structured execution engine.

Living Dashboard

The dashboard is a React app in the company filesystem that agents edit in real time. Custom widgets, custom pages, theme overrides, layout config. Changes appear instantly via HMR. Your agents build your internal tools. No Retool. No deployment. No drag-and-drop.

Zero Infrastructure

One Bun process. One SQLite file. No Docker, no Postgres, no Redis, no vector DB. Install and run. Company config is YAML/Markdown on disk; runtime data (tasks, messages, activity) lives in SQLite with FTS5 full-text search and libSQL native vectors (DiskANN).

Human Gates

Explicit approval points for merge, deploy, spend, and publish -- defined in workflow YAML, not bolted-on permissions. Hardcoded deny patterns prevent agents from touching auth files or secrets. Every agent action is a git commit. Your company has version control and an undo button.

Core Concepts

01 -- Hybrid Storage

YAML for configuration, Markdown for knowledge, SQLite for runtime data. Company config (agents, workflows, skills, knowledge) lives in files you can read, grep, back up with cp, and fork with git clone. Tasks, messages, and activity are stored in SQLite (.data/autopilot.db) with FTS5 full-text search and libSQL native vectors (DiskANN) for fast indexed queries.

02 -- Agents as Employees

Each agent is defined in YAML with an ID, role template, filesystem scope, and tool set. The default template ships 8 agents (CEO, Sam, Alex, Max, Riley, Ops, Morgan, Jordan) across 8 roles (meta, strategist, planner, developer, reviewer, devops, marketing, design). Each agent runs as a TanStack AI session via the TanStack AI with a 6-layer context assembly (identity + company state + agent memory + task context + skills discovery + tool list).

03 -- Tools as Actions

Agents don't produce text output. They call unified tools -- typed function calls with clear targets and effects. task(), message(), pin(), search_index(), fetch(), web_search(). Agent thinking is internal. Only tool calls produce observable results.

04 -- Workflows as Files

Workflows define how work moves through the company. They are YAML files in team/workflows/. Human gates are first-class workflow steps, not afterthought permissions. The CEO agent owns workflow definitions; any agent can propose changes with evidence.

05 -- Orchestrator

A single Bun process that watches the filesystem for changes, matches events against workflow rules, spawns agent sessions, routes tasks, runs cron schedules, handles webhooks, and captures session streams for observability. It is the only long-running process -- agents are ephemeral.

Quick Start

bun add -g @questpie/autopilot
autopilot init my-company
cd my-company
# If using an API key (not needed with OpenRouter API key):
export OPENROUTER_API_KEY=sk-or-xxx

autopilot start
autopilot chat ceo "Build a landing page for our product"
autopilot attach max

What you need: Bun runtime + Anthropic API key or OpenRouter API key. That's it.

Documentation

• Getting Started -- Prerequisites, installation, first intent, first attach
• Tools -- The 12 unified tools. Why not chat. Code examples
• Architecture -- Six-layer stack, zero-infra design, orchestrator internals
• Agents -- Agent definitions, role templates, memory isolation, context assembly
• Workflows -- YAML workflows, human gates, CEO ownership
• Context & Memory -- 6-layer context assembly, persistent memory, memory extraction
• Living Dashboard -- Widgets, theme overrides, layout config, custom pages
• Skills -- Markdown knowledge packages, agentskills.io format, customizable templates
• Artifacts -- Agent-created previews, artifact router, cold-start, port pool
• Features -- Unified search, embeddings, multiple providers, transport plugins
• Integrations -- 3-part pattern, secret management, GitHub, Linear, Slack, Stripe
• Use Cases -- Dev shop, self-building tools, infrastructure management
• CLI Reference -- All Autopilot commands