Install

curl -fsSL https://raw.githubusercontent.com/launchapp-dev/ao/main/install.sh | bash

options

# Specific version
AO_VERSION=v0.0.11 curl -fsSL https://raw.githubusercontent.com/launchapp-dev/ao/main/install.sh | bash

# Custom directory
AO_INSTALL_DIR=/usr/local/bin curl -fsSL https://raw.githubusercontent.com/launchapp-dev/ao/main/install.sh | bash

prerequisites

You need at least one AI coding CLI:

npm install -g @anthropic-ai/claude-code    # Claude (recommended)
npm install -g @openai/codex                # Codex
npm install -g @google/gemini-cli           # Gemini

---

What is AO?

AO turns a single YAML file into an autonomous software delivery pipeline.

You define agents, wire them into phases, compose phases into workflows, schedule everything with cron — and AO's daemon handles the rest: dispatching tasks to AI agents in isolated git worktrees, managing quality gates, and merging the results.

                ┌──────────────────────────────────────────────────┐
                │               AO Daemon (Rust)                   │
                │                                                  │
  ┌────────┐    │    ┌───────────┐    ┌───────────┐    ┌────────┐ │    ┌────────┐
  │ Tasks  │───▶│───▶│  Dispatch │───▶│  Agents   │───▶│ Phases │─│──▶│  PRs   │
  │        │    │    │  Queue    │    │           │    │        │ │    │        │
  │ TASK-1 │    │    │ priority  │    │ Claude    │    │ impl   │ │    │ PR #42 │
  │ TASK-2 │    │    │ routing   │    │ Codex     │    │ review │ │    │ PR #43 │
  │ TASK-3 │    │    │ capacity  │    │ Gemini    │    │ test   │ │    │ PR #44 │
  └────────┘    │    └───────────┘    └───────────┘    └────────┘ │    └────────┘
                │                                                  │
                │    Schedules: work-planner (5m), pr-reviewer     │
                │    (5m), reconciler (5m), PO scans (2-8h)        │
                └──────────────────────────────────────────────────┘

---

Quick Start

cd your-project                          # any git repo
ao doctor                                # check prerequisites
ao setup                                 # initialize .ao/

ao task create --title "Add rate limiting" --type feature --priority high
ao workflow run --task-id TASK-001        # run once

ao daemon start --autonomous             # or go fully autonomous

---

Everything in One YAML

Agents Bind models, tools, MCP servers, and system prompts to named profiles. Route by task complexity. agents: default: model: claude-sonnet-4-6 tool: claude mcp_servers: ["ao", "context7"] work-planner: system_prompt: | Scan tasks, check dependencies, enqueue ready work for the daemon. model: claude-sonnet-4-6 tool: claude	Phases Reusable execution units. Three modes: agent (AI with decision contracts), command (shell), manual (human gate). phases: implementation: mode: agent agent: default directive: "Implement production code." decision_contract: min_confidence: 0.7 max_risk: medium push-branch: mode: command command: program: git args: ["push", "-u", "origin", "HEAD"]
Workflows Compose phases into pipelines with skip conditions and post-success hooks. workflows: - id: standard phases: - requirements - implementation - push-branch - create-pr post_success: merge: strategy: squash auto_merge: true cleanup_worktree: true	Schedules Cron-based autonomous execution. The daemon runs your workflows on a cadence. schedules: - id: work-planner cron: "*/5 * * * *" workflow_ref: work-planner enabled: true - id: pr-reviewer cron: "*/5 * * * *" workflow_ref: pr-reviewer enabled: true

---

The Full Agent Team

AO doesn't run one agent. It runs an entire product organization:

  ┌─────────────────────────────────────────────────────────────────┐
  │                                                                 │
  │   Planners               Builders              Reviewers        │
  │   ╭──────────────╮       ╭──────────────╮       ╭──────────────╮│
  │   │ Work Planner │       │ Claude Eng   │       │ PR Reviewer  ││
  │   │ Reconciler   │       │ Codex Eng    │       │ PO Reviewer  ││
  │   │ Triager      │       │ Gemini Eng   │       │ Code Review  ││
  │   │ Req Refiner  │       │ GLM Eng      │       │              ││
  │   ╰──────────────╯       ╰──────────────╯       ╰──────────────╯│
  │                                                                 │
  │   Product Owners         Architects             Operations      │
  │   ╭──────────────╮       ╭──────────────╮       ╭──────────────╮│
  │   │ PO: Web      │       │ Rust Arch    │       │ Sys Monitor  ││
  │   │ PO: MCP      │       │ Infra Arch   │       │ Release Mgr  ││
  │   │ PO: Workflow │       │              │       │ Branch Sync  ││
  │   │ PO: CLI      │       │              │       │ Doc Drift    ││
  │   │ PO: Runner   │       │              │       │ Wf Optimizer ││
  │   ╰──────────────╯       ╰──────────────╯       ╰──────────────╯│
  │                                                                 │
  └─────────────────────────────────────────────────────────────────┘

Key Concepts

Decision Contracts Every agent phase returns a typed verdict: advance, rework, skip, or fail. Rework loops pass the reviewer's feedback back to the implementer. Configurable max_rework_attempts prevents infinite loops.

Model Routing Route tasks to different models by type and complexity. Low-priority bugfixes go to cheap models. Critical architecture tasks go to Opus. The work-planner agent manages this automatically.

Worktree Isolation Every task gets its own git worktree. Agents work in parallel on separate branches without conflicts. Post-success hooks handle merge, cleanup, and PR creation.

Complexity	Type	Model	Why
low	bugfix/chore	GLM-5-Turbo	Cheapest option
medium	feature	Claude Sonnet	Reliable, fast
medium	UI/UX	Gemini 3.1 Pro	Vision + design expertise
high	refactor	Codex GPT-5.3	Strong code understanding
high	architecture	Claude Opus	Maximum quality
critical	any	Claude Opus	No compromises

---

Claude Code Integration

Install AO Skills for deep AO integration inside Claude Code:

git clone https://github.com/launchapp-dev/ao-skills.git ~/ao-skills
claude --plugin-dir ~/ao-skills

Slash Commands Command What it does /setup-ao Initialize AO in your project /getting-started Install, concepts, first task /workflow-authoring Write custom YAML workflows /pack-authoring Build workflow packs /mcp-setup Connect AI tools via MCP /troubleshooting Debug common issues

Auto-Loaded References Skill Coverage configuration Config files, state layout, model routing task-management Full task lifecycle via CLI and MCP daemon-operations Daemon monitoring and troubleshooting workflow-patterns Patterns from 150+ autonomous PRs agent-personas PO, architect, auditor agents mcp-tools Complete ao.* tool reference

---

CLI

ao task          Create, list, update, prioritize tasks
ao workflow      Run and manage multi-phase workflows
ao daemon        Start/stop the autonomous scheduler
ao queue         Inspect and manage the dispatch queue
ao agent         Control agent runner processes
ao output        Stream and inspect agent output
ao doctor        Health checks and auto-remediation
ao setup         Interactive project initialization
ao requirements  Manage product requirements
ao mcp           Start AO as an MCP server
ao web           Launch the embedded web dashboard
ao status        Project overview at a glance

---

Architecture

108k lines of Rust · 16 crates

orchestrator-cli ·········· 30k    CLI commands and dispatch
orchestrator-core ·········· 18k   Services, state, config
orchestrator-config ········ 13k   Workflow YAML compilation
workflow-runner-v2 ········· 9k    Phase execution engine
agent-runner ··············· 6k    LLM CLI process management
llm-cli-wrapper ············ 6k    CLI tool abstraction layer
orchestrator-web-server ···· 5k    Embedded React dashboard
protocol ··················· 5k    Shared types, model routing
orchestrator-daemon ········ 4k    Daemon scheduler runtime
oai-runner ················· 3k    OpenAI-compatible API client
orchestrator-providers ····· 3k    Provider abstractions
orchestrator-web-api ······· 3k    GraphQL API layer
+ 4 supporting crates

graph LR
    A[CLI] --> B[Core Services]
    A --> C[Daemon Runtime]
    B --> D[Workflow Runner]
    D --> E[Agent Runner]
    E --> F[LLM CLI Wrapper]
    F --> G[claude / codex / gemini]
    B --> H[Config]
    H --> I[YAML Compiler]
    A --> J[Web Server]
    J --> K[GraphQL API]
    K --> B
    C --> D
    style A fill:#1f6feb,stroke:#1f6feb,color:#fff
    style C fill:#1f6feb,stroke:#1f6feb,color:#fff
    style J fill:#1f6feb,stroke:#1f6feb,color:#fff

Loading

---

Platforms

Platform	Architecture
macOS	Apple Silicon (M1+)	aarch64-apple-darwin
macOS	Intel	x86_64-apple-darwin
Linux	x86_64	x86_64-unknown-linux-gnu
Windows	x86_64	x86_64-pc-windows-msvc

---

Update

curl -fsSL https://raw.githubusercontent.com/launchapp-dev/ao/main/install.sh | bash

Uninstall

rm -f ~/.local/bin/ao ~/.local/bin/agent-runner ~/.local/bin/llm-cli-wrapper

_{Built with Rust. Powered by AI. Ships code autonomously.}