EZ AI Pipeline

A production-ready CLI tool for executing prompt optimization pipelines using AI models. Features a beautiful terminal interface, multi-stage pipeline execution, cost-efficient model selection, and a built-in web dashboard.

Features

• Multi-Stage Pipelines: Define complex workflows with multiple AI processing stages
• Conditional Execution: Skip stages, run conditionally, or exit early based on results
• Cost Tracking: Automatic token counting and cost calculation for all API calls
• Beautiful Terminal UI: Progress indicators, spinners, and formatted output
• Variable Interpolation: Pass data between stages using template variables
• Multiple Output Formats: JSON, Markdown, and text parsing support
• Pause/Resume/Cancel: Control pipeline execution in real-time
• Web Dashboard: Built-in web viewer to browse pipeline outputs
• Included Pipelines: Ships with ready-to-use pipelines (prompt-optimizer, pipeline-generator)

Installation

# Install globally from npm
npm install -g @ez-corp/ez-ai-pipeline

# Or use with npx
npx @ez-corp/ez-ai-pipeline --help

Quick Start

1. Set up your API key (or use CLI mode)

# Option A: Use your Anthropic API key
export ANTHROPIC_API_KEY="your-api-key"

# Option B: Use CLI mode with Claude Code (no API key needed)
# Just have Claude Code installed and authenticated

2. Run an included pipeline

ez-ai-pipeline run -p prompt-optimizer -i "Write a blog post about productivity"

3. View results in the web dashboard

ez-ai-pipeline web

4. Create your own pipeline

ez-ai-pipeline init my-pipeline

Commands

run - Execute a pipeline

ez-ai-pipeline run --pipeline <id> --input <text>
ez-ai-pipeline run --pipeline <id> --input-file <path>
ez-ai-pipeline run --pipeline <id> --input <text> --output ./results/
ez-ai-pipeline run --pipeline <id> --input <text> --json  # JSON output only
ez-ai-pipeline run --pipeline <id> --input <text> --quiet # Minimal output

list - Show available pipelines

ez-ai-pipeline list
ez-ai-pipeline list --json

validate - Validate a pipeline configuration

ez-ai-pipeline validate --pipeline <id>

init - Create a new pipeline from template

ez-ai-pipeline init <name>

web - Start the web dashboard

ez-ai-pipeline web              # Start on default port 3000
ez-ai-pipeline web -p 8080      # Start on custom port

Opens a local web interface to browse pipelines and execute them directly in your browser.

Features:

• Run Pipelines: Execute any pipeline with real-time progress tracking
• Live Progress: Watch stages complete with WebSocket-powered updates
• Cost Tracking: See costs accumulate as each stage finishes
• View Outputs: Browse all pipeline execution results
• Stage Details: Explore stage-by-stage outputs and configurations
• CLI Mode: Use Claude Code, OpenCode, or Aider instead of API keys

Execution Modes:

• API Mode: Uses your Anthropic API key (default when ANTHROPIC_API_KEY is set)
• CLI Mode: Shells out to Claude Code, OpenCode, or Aider (no API key required)

The dashboard auto-detects installed CLI tools and lets you choose your preferred execution mode.

Running Pipelines from the Web:

1. Start the web dashboard: ez-ai-pipeline web
2. Navigate to a pipeline from the sidebar
3. Select execution mode (API or CLI) and tool if using CLI mode
4. Enter your prompt in the textarea
5. Click "Run Pipeline" or press Ctrl+Enter
6. Watch real-time progress as each stage executes
7. View the final optimized output

Keyboard Shortcuts:

• Ctrl+Enter - Run the pipeline
• Escape - Cancel a running execution

Included Pipelines

The package ships with ready-to-use pipelines:

prompt-optimizer

Optimizes prompts for better AI responses. Analyzes input, identifies improvements, and generates an enhanced version.

ez-ai-pipeline run -p prompt-optimizer -i "Your prompt to optimize"

pipeline-generator

Generates new pipeline configurations from natural language descriptions.

ez-ai-pipeline run -p pipeline-generator -i "Create a pipeline that summarizes articles"

Pipeline Configuration

Pipelines are TypeScript files in the pipelines/ directory:

import type { PipelineConfig } from "../src/typings/pipeline.ts";

const config: PipelineConfig = {
  pipeline: {
    id: "my-pipeline",
    name: "My Pipeline",
    version: "1.0.0",
    description: "Description of what this pipeline does",
    settings: {
      enableCaching: true,
      enableEarlyExit: true,
      maxRetries: 2,
      timeoutMs: 300000,
    },
  },
  stages: [
    {
      id: "stage-1",
      name: "Analyze Input",
      type: "analyze",
      model: {
        providerID: "anthropic",
        modelID: "claude-haiku-4-5-20251001",
        tier: "small",
        maxTokens: 1024,
        temperature: 0.1,
      },
      prompt: {
        systemPrompt: "You are an expert analyst.",
        template: `Analyze this: {{input_prompt}}`,
        variables: [
          { name: "input_prompt", source: "input", path: "prompt", required: true },
        ],
      },
      output: {
        format: "json",
        extract: [
          { name: "result", path: "result", required: true },
        ],
      },
    },
  ],
};

export default config;

Stage Types

• analyze - Analyze and extract information
• structure - Restructure or organize content
• enhance - Improve or enrich content
• validate - Validate results against criteria
• test - Test outputs or run checks
• iterate - Make iterative improvements
• custom - Custom processing stage

Variable Sources

Variables can be sourced from:

• input - The original input to the pipeline
• previousStage - Output from a previous stage
• context - Accumulated context throughout the pipeline
• static - Static values defined in the configuration

Conditions

Skip If

Skip a stage if a condition is met:

conditions: {
  skipIf: {
    condition: { path: "status", equals: "complete" },
    sourceStage: "stage-1",
  },
}

Run If

Only run a stage if a condition is met:

conditions: {
  runIf: {
    condition: { path: "needsWork", equals: true },
    sourceStage: "stage-1",
  },
}

Early Exit

Exit the pipeline early if a condition is met:

conditions: {
  earlyExit: {
    condition: { path: "overall", equals: "COMPLETE" },
    returnStage: "stage-1",
  },
}

Output

Results are saved to the outputs/ directory:

• result-{pipeline}-{timestamp}.json - Full pipeline results
• prompt-{pipeline}-{timestamp}.md - Optimized prompt (for prompt pipelines)

Web API

The web dashboard exposes a REST API and WebSocket endpoint for programmatic access:

REST Endpoints

Running a Pipeline via API

# Start execution
curl -X POST http://localhost:3000/api/pipelines/prompt-optimizer/run \
  -H "Content-Type: application/json" \
  -d '{"input": "Your prompt to optimize"}'

# Response: { "executionId": "exec_123...", "pipelineId": "prompt-optimizer", "status": "started" }

# Check status
curl http://localhost:3000/api/executions/exec_123...

# Cancel execution
curl -X POST http://localhost:3000/api/executions/exec_123.../cancel

WebSocket for Real-Time Updates

Connect to ws://localhost:3000/ws for real-time execution progress:

const ws = new WebSocket('ws://localhost:3000/ws');

// Subscribe to an execution
ws.send(JSON.stringify({ type: 'subscribe', executionId: 'exec_123...' }));

// Receive events
ws.onmessage = (event) => {
  const data = JSON.parse(event.data);
  // Event types: stage:started, stage:completed, stage:failed,
  //              stage:skipped, execution:completed, execution:cancelled
  console.log(data);
};

Development

# Run tests
bun test

# Type check
bun run typecheck

# Format code
bun run format

Project Structure

ez-ai-pipeline/
├── src/
│   ├── index.ts              # CLI entry point
│   ├── cli/                  # CLI commands and UI
│   │   └── commands/         # Command implementations (run, list, validate, init, web)
│   ├── core/                 # Pipeline execution engine
│   ├── schema/               # Zod schemas and validation
│   ├── typings/              # TypeScript type definitions
│   ├── utils/                # Utilities (cost, templates, etc.)
│   └── config/               # Configuration and constants
├── pipelines/                # Included pipeline definitions
├── web/                      # Web dashboard source
├── web-dist/                 # Built web assets (included in npm package)
├── outputs/                  # Generated outputs
├── tests/                    # Test files
└── package.json

Supported Models

Anthropic

• claude-opus-4-5-20250929 (large)
• claude-sonnet-4-5-20250929 (medium)
• claude-haiku-4-5-20251001 (small)

OpenAI

• gpt-4o (medium)
• gpt-4o-mini (small)
• gpt-4-turbo (large)

Google

• gemini-1.5-pro (large)
• gemini-1.5-flash (small)
• gemini-2.0-flash (small)

License

MIT License - see LICENSE for details.

Contributing

See CONTRIBUTING.md for contribution guidelines.