Workflow Documentation

MuAPI’s workflow system allows you to build sophisticated, multi-stage AI pipelines using a visual node-based editor or natural language chat.

Core Concepts

1. Nodes and Categories

Each node in a workflow represents a specific AI model or utility.

  • Text: LLM processing, prompt engineering.
  • Image: Generation, upscaling, editing.
  • Video: Motion, effects, high-fidelity generation.
  • Audio: Music generation, sound effects.
  • Utility: Logic nodes like "Passthrough" or "Concatenator".
  • API: Direct access to third-party integrations (Straico, WaveSpeed).

2. Edges and Connections

Connections (edges) define the flow of data.

  • Handles: Nodes have specific input/output handles (e.g., imageOutput -> videoInput).
  • Dynamic References: Use Jinja2 syntax to inject data: {{ node_id.outputs[0].value }}.

Agentic Workflow Architect

Building complex graphs is easy with the built-in AI assistant.

  • Natural Language Creation: "Design a marketing pipeline that starts with a text prompt, creates a high-res image, and then generates a 5s video."
  • To-and-Fro Planning: For broad requests, the architect will lead a planning discussion, proposing multiple architectural options before built the graph.
  • Intelligent Refinement: Ask the architect to "Change the video model to Runway" or "Add a background removal step".

Running Workflows

Manual Execution

  • Run button: Initiates the complete graph from the start nodes.
  • Node-level Run: Test individual nodes in isolation with custom parameter overrides.

API Orchestration

Workflows can be triggered via REST API.

  • Endpoint: POST /api/workflow/{workflow_id}/run
  • Webhook Support: Provide a webhook_url to receive a single notification when the entire workflow completes, or granular updates for each node.

CLI Orchestration & LLM Discovery

Workflows are fully integrated with the muapi-cli to provide an LLM-friendly interface for AI agents (like Claude Code, Cursor, etc.).

  • Discovery: Agents can fetch a complete catalog of available workflows and their textual descriptions: 
    muapi workflow discover --output-json
  • Analysis: Once an agent chooses a workflow, it can retrieve its structural schema: 
    muapi workflow get <workflow_id> --output-json
  • Interactive Execution: For a guided CLI experience, use the interactive run tool which prompts for all required inputs: 
    muapi workflow run-interactive <workflow_id>
  • Create & Edit: You can prompt the AI architect directly from the terminal to build new functionality or edit existing workflows: 
    muapi workflow create "take a text prompt, generate an image with flux-dev, then upscale it" --view

External API specialized Integration

Access specialized cinematic and multimodal models via integrated helpers:

  • Straico: One interface for dozens of leading AI models.
  • WaveSpeed: Optimized for high-speed cinematic generation with automated spec parsing.