AWL is an open-source framework and curated library of intelligent, agent-driven scientific workflows. We are building a future where every bioinformatics workflow is not just executable—but agentic: aware of its purpose, capable of answering questions, and ready to collaborate.

✨ Built by and for the bioinformatics and AI community. 🧜‍♂️ Focused on reproducible, intelligent, and scalable analysis.

AWL is both a framework and a collection of reusable agents designed to modernize scientific workflows through automation, intelligence, and collaboration.

It consists of:

• 🧱 A modular agent framework (runtime, CLI, schema, dev tools)
• 🤖 Prebuilt domain-specific agents (e.g., RNA-seq, SV-calling)
• 📚 A growing community catalog of shareable workflows and tools
• ↻ A human-in-the-loop design: agents that work with researchers

Think: the power of CWL or Nextflow, but with an LLM-aware assistant behind every step.

Traditional workflows are rigid and opaque. Agentic workflows are flexible, self-documenting, and interactive.

AWL agents:

• Understand their domain (e.g., RNA-seq, GWAS, variant calling)
• Know their tools (via CWL, Docker, environment metadata)
• Respond to prompts (LLM integration)
• Validate inputs, run tests, and explain decisions
• Coordinate across projects with a project manager agent

agentic-workflow-language/
├── awl-core/          # The runtime engine and orchestration layer
├── awl-cli/           # Command-line tool for managing agents
├── awl-spec/          # YAML schemas and AWL definitions
├── awl-catalog/       # Shared library of agent metadata and tools
├── awl-handbook/      # Community docs and onboarding
├── sv-agent/          # Example: agent for structural variant workflows
├── rnaseq-agent/      # Example: agent for RNA-seq pipelines
└── example-workflows/ # Publicly available test pipelines

# Install CLI (coming soon)
pip install awl

# Clone an example agent
git clone https://github.com/agentic-workflow-language/sv-agent.git

# Run interactively
awl run sv-agent --chat

# Or run headlessly
awl run sv-agent --input config.yaml

AWL leverages and extends:

• Common Workflow Language (CWL)
• Docker and Conda environments
• JSON/YAML schema validation
• LLMs like GPT-4, Claude, and Ollama for interpretability
• Vector databases (optional) for retrieval-augmented prompting

We believe bioinformatics workflows should be:

• Reusable
• Understandable
• Auditable
• Promptable

Join us in building agentic infrastructure for the next generation of scientific computing.

📬 hello@awl.is
🌐 awl.is
📣 GitHub Discussions (coming soon)
💬 Chat with us via Matrix (coming soon)

We welcome contributions from researchers, developers, and toolmakers. Whether you're wrapping CWL tools, building new agents, writing docs, or just testing things—your help is valuable.

Start here:

git clone https://github.com/agentic-workflow-language/awl-handbook.git
cd awl-handbook

See CONTRIBUTING.md for guidelines.

AWL is released under the Apache 2.0 License. Open, extensible, and free for all.

We believe in a future where:

• Scientists collaborate with their workflows like they would with a lab technician
• Agents evolve with new data, methods, and tools
• Workflows are not just pipelines—but peers in the discovery process

Let’s build that future together.

— The AWL Team