<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> npcsh is a composable multi-agent shell designed for building and running AI teams directly from the command line. It empowers users to orchestrate complex AI workflows by leveraging multi-modal LLMs, slash commands, and interactive modes. The shell facilitates agent scheduling, context engineering, and the creation of custom execution templates (Jinxes), offering a powerful interface for collaborative AI development and task execution.

How It Works

The core of npcsh is its agentic shell architecture, enabling users to define and manage teams of AI agents. It supports flexible agent definition formats, including YAML (.npc files), Markdown (agents.md), and individual files (agents/). npcsh integrates with a wide array of LLM providers through LiteLLM, allowing for diverse model utilization. Key features include Jinxes for templated workflows, NQL for AI-augmented database querying, knowledge graph construction, and multi-agent discussion capabilities, providing a robust framework for AI task orchestration.

Quick Start & Requirements

• Installation: pip install 'npcsh[lite]'
• Execution: Run npcsh to enter the shell.
• Prerequisites: Python, espeak, portaudio19-dev, python3-pyaudio, ffmpeg, libcairo2-dev, libgirepository1.0-dev (Linux); brew install portaudio ffmpeg pygobject3 (macOS); Ollama for local models; API keys for cloud providers (e.g., OpenAI, Anthropic, Gemini) configured via .env.
• Documentation: Full documentation is available at npc-shell.readthedocs.io.

Highlighted Details

• Composable command-line shell for orchestrating AI agent teams.
• Supports multiple agent definition formats (.npc YAML, agents.md, agents/) and integrates with numerous LLM providers via LiteLLM.
• Features include NQL for AI-SQL queries, Knowledge Graphs, Jinxes for templated workflows, and MCP integration.
• Enables image, audio, and video generation, alongside GUI automation with vision capabilities.

Maintenance & Community

• Community: Active Discord server available for support and discussion.
• Contributions: Contributions are welcomed via GitHub issues and pull requests.
• Resources: Links to research papers, a newsletter, and consulting services are provided.

Licensing & Compatibility

• License: MIT License, permitting commercial use and integration with closed-source projects.
• Compatibility: Agents from other ecosystems (e.g., Claude Code, Codex, Amp) can be integrated by placing their definitions in the npc_team/ directory.

Limitations & Caveats

The native Rust edition is explicitly marked as experimental. Benchmark comparisons with other agent coders are pending ("coming soon").