Deer Workflow

Deer Workflow is a code-first runtime for coordinating AI agents through reusable execution graphs. Instead of hiding planning and recovery inside a long agent conversation, it lets developers express phases, branching, parallel work, inputs, and failure handling in TypeScript. Individual nodes can then assign semantic tasks to supported coding-agent runtimes.

Highlights

• Generates runnable workflow modules from natural-language requirements through a bundled creator skill
• Supports Codex by default, with built-in adapters for Claude Code and Pi
• Runs independent tasks in parallel and combines their results in later phases
• Provides a live terminal interface for interactive work and JSONL events for servers or CI pipelines
• Includes documented examples for deep research and structured article production

Deer Workflow is aimed at developers who want agent automation that can be reviewed, tested, versioned, and integrated with existing code. It is especially useful for research pipelines, content systems, and multi-stage engineering tasks where execution must remain understandable. The project is distributed under the MIT license and can be installed or run locally at no charge, although the connected agent providers may have their own usage costs.