Skip to content

CLAUDE.md

Latest commit

 

History

History
61 lines (46 loc) · 1.77 KB

CLAUDE.md

File metadata and controls

61 lines (46 loc) · 1.77 KB

CLAUDE.md

Code Style

  • AVOID inline comments

Commands

pip install -e .[dev]       # Install dev dependencies
pytest                      # Run tests
ruff check --fix --select I && ruff format  # Lint and format

Architecture

DAG-based workflow library connecting Gradio Spaces, ML models, and Python functions.

Core Files

  • graph.py: Graph class - holds nodes, edges, launches server
  • node.py: GradioNode, FnNode, InferenceNode, InteractionNode
  • port.py: Port, ScatteredPort, GatheredPort for parallel execution
  • edge.py: Edge class for node connections
  • executor.py: Executes workflow nodes
  • server.py: FastAPI server with WebSocket for real-time UI updates
  • state.py: Session state persistence with SQLite
  • frontend/: Svelte frontend for workflow visualization

Node Definition Pattern

Nodes use inputs and outputs dicts:

node = GradioNode(
    space_or_url="owner/space",
    api_name="/endpoint",
    inputs={
        "param": gr.Textbox(label="Label"),  # UI input
        "other": other_node.output_port,      # Port connection
        "fixed": "constant_value",            # Fixed value
    },
    outputs={
        "result": gr.Audio(label="Result"),
        "hidden": gr.Text(visible=False),     # Hidden output
    },
)

Key Patterns

  1. Port access: node.port_name returns a Port for connections
  2. Graph creation: Graph(nodes=[...]) auto-wires from port connections in inputs
  3. Internal attrs: Use _ prefix (_name, _fn) to avoid port name conflicts

Issue Workflow

  1. Create branch: git checkout -b fix-issue-NUMBER
  2. Implement fix
  3. Run: pytest && ruff check --fix --select I && ruff format
  4. Do not commit - leave for user