lider 0.1.0

A lightweight framework for building and running AI multi-agent workflows with a terminal UI

lider

A lightweight framework for building and running AI multi-agent workflows with a terminal UI.

Installation

Requires Python 3.11+.

# Install uv if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies and the project
uv sync

# Install dev dependencies (for testing)
uv sync --group dev

Usage

lider path/to/workflow.py

Workflows are plain Python files with an async def run() function.

SDK

from lider.sdk import agent_node, input_node, process_node

async def run():
    question = await input_node("Ask something:")
    answer = await agent_node(question, tool="claude")

    # Run an OS process and chain its output to an agent
    logs = await process_node("journalctl -n 50", shell=True)
    summary = await agent_node(logs, tool="claude")

• input_node(prompt) — prompts the user for input
• agent_node(prompt, tool=...) — sends a prompt to an AI model ("claude" or "gemini")
• process_node(command, cwd=None, shell=False) — runs an OS process, streams output to the TUI, and returns the full stdout as a string

Currently supported tools: Claude Code CLI and Gemini CLI.

Examples

debate.py

lider example-workflows/debate.py

Asks a question, gets answers from Claude and Gemini in parallel, then synthesizes them into a single response.

multi-model-code-review.py

lider example-workflows/multi-model-code-review.py

Runs a code review in parallel with Claude and Gemini, then synthesizes the two reviews into a prioritized, authoritative report with a final verdict.

new-feature-pipeline.py

lider example-workflows/new-feature-pipeline.py

End-to-end feature development pipeline: creates a git worktree, drafts a spec interactively with Claude, generates an implementation plan, implements it, runs acceptance and code reviews in parallel, fixes findings, and pushes the branch.