https://github.com/flying-dice/abtree

Behaviour trees for AI agents. Define agentic workflows in YAML, drive deterministic agent runs from a CLI works with any agent, Claude, Codex, Copilot, Gemini etc.
https://github.com/flying-dice/abtree

agent-framework agent-orchestration agentic-ai agentic-workflows ai-agents autonomous-agents behavior-tree claude claude-code codex copilot gemini llm llm-agents

Last synced: 2 days ago
JSON representation

Behaviour trees for AI agents. Define agentic workflows in YAML, drive deterministic agent runs from a CLI works with any agent, Claude, Codex, Copilot, Gemini etc.

• Host: GitHub
• URL: https://github.com/flying-dice/abtree
• Owner: flying-dice
• Created: 2026-05-08T19:42:24.000Z (24 days ago)
• Default Branch: main
• Last Pushed: 2026-05-09T23:00:32.000Z (23 days ago)
• Last Synced: 2026-05-10T00:36:20.804Z (23 days ago)
• Topics: agent-framework, agent-orchestration, agentic-ai, agentic-workflows, ai-agents, autonomous-agents, behavior-tree, claude, claude-code, codex, copilot, gemini, llm, llm-agents
• Language: TypeScript
• Homepage: http://abtree.sh/
• Size: 586 KB
• Stars: 1
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          


  



abtree




  Hoping. Behaving.


  Treat agent instructions like the software they are. Clear steps, predictable behavior, real answers when something goes wrong.





  Docs ·

  Get started ·

  How it works



---

## What it does

abtree is a runtime for agent workflows. Author a tree as JSON, YAML, or compile it from the TypeScript DSL. Ship it through any transport your team already uses — abtree never sees the distribution; it only reads the file at the path you point it at. Your agent drives execution through three commands (`next`, `eval`, `submit`) and only ever sees the next step.

![A behaviour tree walking the SRP refactor loop — one failed pass, one clean retry](docs/public/example.svg)

The trace above is a `Refactor_Loop` sequence with three actions. Green nodes ran and succeeded, red ran and failed, the pink ring marks the cursor. The runtime regenerates the diagram after every state change, so what the agent did and what it skipped is on disk by the time it finishes.

## Install

**macOS / Linux**

```sh

curl -fsSL https://github.com/flying-dice/abtree/releases/latest/download/install.sh | sh

```

**Windows (PowerShell)**

```powershell

irm https://github.com/flying-dice/abtree/releases/latest/download/install.ps1 | iex

```

## Read the docs

Concepts, guides, CLI reference, and a five-minute walkthrough all live at **[abtree.sh](https://abtree.sh)**.

→ [**Get started in five minutes**](https://abtree.sh/getting-started)