AI agent pipeline orchestration CLI with interactive TUI, built with Rust for performance and distributed via npm for convenience.
- Multi-Agent Pipelines: Orchestrate multiple AI agents (Claude, Gemini, Cursor) in sequential workflows
- Interactive TUI: Real-time process monitoring with dashboard and detail views built with ratatui
- Slash Commands: Execute commands with autocomplete (
/start,/pause,/resume,/kill,/list) - Event-Driven: Async communication between core engine and UI for responsive interactions
- Cross-Platform: Native binaries for macOS, Linux, Windows (x64 and ARM64)
- Git Integration: Optional git commit automation after pipeline completion
- Human Review Points: Pause pipelines for manual review before continuing
# Install globally via npm
npm install -g pipeline-kit
# Or use directly with npx
npx pipeline-kitThe easiest way to get started is with the init command:
# Initialize .pipeline-kit directory with templates
pipeline-kit init
# Or use minimal templates (1 agent, 1 pipeline)
pipeline-kit init --minimal
# Initialize in a specific directory
pipeline-kit init --path ~/my-projectThis will create:
.pipeline-kit/config.toml- Global configuration.pipeline-kit/agents/- Agent definitions (developer, reviewer).pipeline-kit/pipelines/- Pipeline workflows (simple-task, code-review)
If you prefer manual setup instead of using init:
- Create configuration directory:
mkdir -p .pipeline-kit/{agents,pipelines}- Configure an agent (
.pipeline-kit/agents/developer.md):
---
name: developer
description: A helpful coding assistant
model: claude-sonnet-4.5
color: blue
---
You are a helpful coding assistant. Your goal is to help users with their programming tasks efficiently and accurately.- Create a pipeline (
.pipeline-kit/pipelines/code-review.yaml):
name: code-review
master:
model: claude-sonnet-4.5
system-prompt: |
You are the master orchestrator for a code review pipeline.
Coordinate between agents to ensure high-quality code.
process:
- developer
- reviewer
- HUMAN_REVIEW
sub-agents:
- developer
- reviewer- Set up API keys:
# For Claude
export ANTHROPIC_API_KEY=your_api_key
# For Gemini
export GEMINI_API_KEY=your_api_key
# For Cursor
export CURSOR_API_KEY=your_api_key- Launch the TUI:
pipeline-kit- Start a pipeline:
Type
/start simple-taskin the command input and press Enter.
your-project/
├── .pipeline-kit/
│ ├── config.toml # Global settings
│ ├── agents/ # Agent definitions
│ │ ├── developer.md
│ │ ├── reviewer.md
│ │ └── researcher.md
│ └── pipelines/ # Pipeline workflows
│ ├── code-review.yaml
│ ├── feature-dev.yaml
│ └── bug-fix.yaml
# Enable git integration
git = true
# Default timeout for agent execution (seconds)
timeout = 300Agents are defined in Markdown files with YAML frontmatter:
---
name: agent-name
description: Brief description
model: claude-sonnet-4.5 # or gemini-1.5-pro, cursor-default
color: blue # UI color: blue, green, yellow, red, etc.
---
System prompt for the agent goes here.
You can use multiple paragraphs and markdown formatting.
Key responsibilities:
- Task 1
- Task 2Supported Models:
- Claude:
claude-sonnet-4.5,claude-opus-4,claude-haiku-4 - Gemini:
gemini-1.5-pro,gemini-1.5-flash - Cursor:
cursor-default
Pipelines are defined in YAML:
name: pipeline-name
# Master agent coordinates the workflow
master:
model: claude-sonnet-4.5
system-prompt: |
Your role as the master orchestrator.
Coordinate between sub-agents to achieve the goal.
# Sequential process steps
process:
- researcher # First sub-agent
- developer # Second sub-agent
- HUMAN_REVIEW # Pause for manual review
- reviewer # Final sub-agent
# List of sub-agents used in this pipeline
sub-agents:
- researcher
- developer
- reviewer
# Optional: Required reference files
required-reference-file: falseSpecial Keywords:
HUMAN_REVIEW: Pauses the pipeline for manual review. Resume with/resume <process-id>
Launch the interactive terminal UI:
pipeline-kitKeyboard Controls:
↑/↓orj/k: Navigate between processesTab: Autocomplete slash commandsEnter: Execute commandEsc: Clear inputPageUp/PageDown: Fast scroll in detail viewqorCtrl+C: Quit
Execute commands directly from the command line:
# Initialize a new project
pipeline-kit init
pipeline-kit init --minimal
pipeline-kit init --force
# Start a pipeline (TUI mode)
pipeline-kit
# Show help
pipeline-kit --helpAvailable commands in TUI mode:
| Command | Description | Example |
|---|---|---|
/start <name> |
Start a new pipeline | /start code-review |
/pause <id> |
Pause a running process | /pause a1b2c3d4 |
/resume <id> |
Resume a paused process | /resume a1b2c3d4 |
/kill <id> |
Kill a running process | /kill a1b2c3d4 |
/list |
List all active processes | /list |
/detail <id> |
Show process details | /detail a1b2c3d4 |
Tip: Use Tab for command autocomplete and process ID suggestions.
Pipeline Kit supports multiple AI providers through adapters:
export ANTHROPIC_API_KEY=your_api_keyModels:
claude-sonnet-4.5(recommended)claude-opus-4claude-haiku-4
Features: Streaming responses, tool calling, vision support
export GEMINI_API_KEY=your_api_keyModels:
gemini-1.5-progemini-1.5-flash
Features: Streaming responses, multimodal input
export CURSOR_API_KEY=your_api_keyModels:
cursor-default
Features: Code-focused assistant
name: code-review
master:
model: claude-sonnet-4.5
system-prompt: |
Orchestrate a thorough code review process.
Ensure code quality, testing, and documentation.
process:
- developer # Implements the feature
- reviewer # Reviews the code
- HUMAN_REVIEW # Manual approval
sub-agents:
- developer
- reviewername: feature-dev
master:
model: claude-sonnet-4.5
system-prompt: |
Guide the development of a new feature from planning to implementation.
process:
- researcher # Research requirements
- architect # Design the solution
- developer # Implement the feature
- tester # Write tests
- HUMAN_REVIEW # Final review
sub-agents:
- researcher
- architect
- developer
- testername: bug-fix
master:
model: claude-sonnet-4.5
system-prompt: |
Coordinate debugging and fixing process.
process:
- debugger # Identify the issue
- developer # Implement the fix
- tester # Verify the fix
sub-agents:
- debugger
- developer
- testerPipeline Kit uses a monorepo architecture with clear separation of concerns:
┌─────────────────┐
│ User (CLI) │
└────────┬────────┘
│
▼
┌─────────────────────┐
│ TypeScript Wrapper │ (Platform detection & binary launcher)
│ pipeline-kit-cli │
└────────┬────────────┘
│
▼
┌──────────────────────────────────────────┐
│ Rust Binary (pipeline-kit-rs) │
├──────────────────────────────────────────┤
│ ┌────────┐ ┌──────────────┐ │
│ │ TUI │ ◄────────►│ Core │ │
│ │(ratatui)│ Op/Event │ (Business) │ │
│ └────────┘ └───┬──────────┘ │
│ │ │
│ ┌────────┼────────┐ │
│ │ │ │ │
│ ┌───▼──┐ ┌───▼────┐ ┌▼────┐│
│ │State │ │Pipeline│ │Agent││
│ │Mgr │ │Engine │ │Mgr ││
│ └──────┘ └────────┘ └─────┘│
└──────────────────────────────────────────┘
Key Components:
- pk-protocol: Shared data structures and IPC definitions
- pk-core: Business logic (StateManager, PipelineEngine, AgentManager)
- pk-tui: Interactive terminal UI with widgets
- pk-cli: Binary entry point that wires everything together
Want to contribute? See CONTRIBUTING.md for:
- Development setup
- Project structure details
- Testing strategy
- Coding conventions
- PR submission guidelines
# Clone repository
git clone https://github.com/Vooster-AI/pipeline-kit.git
cd pipeline-kit
# Build Rust workspace
cd pipeline-kit-rs
cargo build --release
# Run tests
cargo test --workspace
# Run the TUI
cargo run --release --bin pipelineIf you get "binary not found" error:
# Check vendor directory
ls -la node_modules/pipeline-kit/vendor/
# Reinstall to download binaries
npm install -g pipeline-kit --forceEnsure environment variables are set:
# Check if API keys are set
echo $ANTHROPIC_API_KEY
echo $GEMINI_API_KEY
echo $CURSOR_API_KEY
# Set in your shell profile (~/.bashrc, ~/.zshrc)
export ANTHROPIC_API_KEY=your_key_herePipeline Kit looks for .pipeline-kit/ in:
- Current directory
- Parent directories (up to repository root)
- Home directory (
~/.pipeline-kit/)
Verify your configuration directory exists:
ls -la .pipeline-kit/| Platform | Architecture | Binary Name | Status |
|---|---|---|---|
| macOS | Intel (x64) | macos-x64 | ✅ Supported |
| macOS | Apple Silicon (ARM64) | macos-arm64 | ✅ Supported |
| Linux | x86_64 | linux-x64 | ✅ Supported |
| Linux | ARM64 | linux-arm64 | ✅ Supported |
| Windows | x64 | windows-x64 | ✅ Supported |
| Windows | ARM64 | windows-arm64 |
MIT License - see LICENSE file for details.
Built with reference to the excellent codex-cli architecture.
Made with ❤️ by the Vooster AI team