AI Workflow Builder

# AI Workflow Builder

Chain any combination of muapi.ai generation steps into automated pipelines. The AI architect converts your plain-language description into a runnable node graph.

## Prerequisites

Always ensure the `muapi` CLI is up-to-date before running any commands. Reinstall from source at the start of every session:

```bash
pip install --upgrade /path/to/muapi-cli
```

Replace `/path/to/muapi-cli` with the actual path to the `muapi-cli` directory in the workspace (the directory containing `pyproject.toml`).

## Core Operations

1. **Generate** (`generate-workflow.sh`) — AI architect creates a workflow from a description
2. **Discover** (`discover-workflow.sh`) — Find a relevant existing workflow by natural language
3. **Edit** (`generate-workflow.sh --workflow-id`) — Modify an existing workflow with a prompt
4. **Interactive Run** (`interactive-run.sh`) — Prompt for inputs and execute a workflow
5. **Run** (`run-workflow.sh`) — Execute a workflow, poll node-by-node, collect outputs
6. **CLI** (`muapi workflow`) — Full CRUD + visualization directly from the terminal

---

## Agent Guided Discovery & Selection

As an AI agent, you have the ability to read and understand the purpose of available workflows to select the best one for the user's task (e.g., "create a UGC video").

1. **Discover**: Fetch the catalog of available workflows and their descriptions in JSON format.
   ```bash
   muapi workflow discover --output-json
   ```
2. **Match (Internal Reasoning)**: Use your LLM capabilities to analyze the `name`, `category`, and `description` fields of the returned workflows. Find the best match for the user's intent.
3. **Analyze**: If you find a promising candidate, inspect its structure to ensure it has the necessary nodes and parameters.
   ```bash
   muapi workflow get <workflow_id>
   ```
   **CRITICAL RULE**: The output of `muapi workflow get` will include an "API Inputs" table. You MUST read this table to understand what inputs are required.
4. **Choose & Confirm & Prompt User**:
   - If one workflow is a perfect match, you MUST ask the user to provide the exact values for the required API inputs before executing it. **Never invent or guess input values (like prompts, URLs, etc.) on your own.**
   - If multiple workflows are highly relevant, present the options to the user with their descriptions and ask them to confirm which one to use, and also ask for the required inputs.
   - If no workflow matches the user's complex request, offer to **architect** a new one using `muapi workflow create`.

### Example Agent Reasoning
> "The user wants a product promo video. I fetched the catalog using `discover`. I see two potential workflows:
> 1. `wf_123`: 'Product promo with background music'
> 2. `wf_456`: 'Simple video gen'
> I will analyze `wf_123` with `get`. It has the required nodes. I will suggest `wf_123` or just run it if the match is precise."

---

## Protocol: Building a Workflow

### Step 1 — Describe your pipeline

```bash
muapi workflow create "take a text prompt, generate an image with flux-dev, then upscale it to 4K"
```

The architect returns a workflow with a unique ID and a node graph. Save the ID.

### Step 2 — Inspect and visualize

```bash
# Rich ASCII node graph in the terminal
muapi workflow get <workflow_id>

# Or raw JSON
muapi workflow get <workflow_id> --output-json
```

### Step 3 — Run it

```bash
# Run with specific inputs
muapi workflow execute <workflow_id> \
  --input "node1.prompt=a glowing crystal cave at midnight"

# Use --download to pull results locally
muapi workflow execute <workflow_id> \
  --input "node1.prompt=a sunset" \
  --download ./outputs
```

### Step 4 — Discovery (Optional)
If you want to reuse an existing workflow instead of creating a new one:

```bash
# Search by keywords
muapi workflow discover "ugc video"
```

### Step 5 — Interactive Execution
Run a workflow and have the CLI prompt you for each required input:

```bash
muapi workflow run-interactive <workflow_id>
```

---

## Workflow Examples

### Image Pipelines

```bash
# Text → Image → Upscale
muapi workflow create "take a text prompt, generate with flux-dev, upscale the result"

# Text → Image → Background removal → Product shot
muapi workflow create "generate a product image with hidream, remove background, create professional product shot"
```

### Video Pipelines

```bash
# Text → Video
muapi workflow create "generate a 10-second cinematic video from a text prompt using kling-master"

# Image → Video → Lipsync
muapi workflow create "animate an input image with seedance, then apply lipsync from an audio file"
```

---

## Editing an Existing Workflow

```bash
# Add a step
muapi workflow edit <id> --prompt "add a face-swap step after the image generation"

# Swap a model
muapi workflow edit <id> --prompt "change the video model from kling to veo3"
```

---

## CLI Reference

```bash
# List all your workflows
muapi workflow list

# Browse templates
muapi workflow templates

# Generate new workflow
muapi workflow create "text → flux image → upscale → face swap"

# Visualize a workflow
muapi workflow get <id>

# Execute with inputs
muapi workflow execute <id> --input "node1.prompt=a sunset"

# Monitor a run
muapi workflow status <run_id>

# Get outputs
muapi workflow outputs <run_id> --download ./results

# Edit with AI
muapi workflow edit <id> --prompt "add lipsync at the end"

# Rename / delete
muapi workflow rename <id> --name "Product Pipeline v2"
muapi workflow delete <id>
```

---

## MCP Tools (for AI agents)

| Tool | Description |
|------|-------------|
| `muapi_workflow_list` | List user's workflows |
| `muapi_workflow_create` | AI architect: prompt → workflow |
| `muapi_workflow_get` | Get workflow definition + node graph |
| `muapi_workflow_execute` | Run with specific inputs |
| `muapi_workflow_status` | Node-by-node run status |
| `muapi_workflow_outputs` | Final output URLs |

---

## Constraints

- Workflows can contain any combination of muapi.ai nodes (image, video, audio, enhance, edit)
- Node outputs are automatically wired as inputs to downstream nodes
- `--sync` mode waits up to 120s for generation; use `--async` for complex workflows and poll separately
- Run timeouts: 10 minutes maximum per workflow execution