Npcsh, the multi-agent shell, has now reached 50 stars on GitHub (github.com)

Npcsh, a command-line multi-agent shell, has hit 50 stars on GitHub — a modest milestone that highlights growing interest in developer-centric tooling for agentic workflows. Built around an “NPC Data Layer,” npcsh extends a familiar bash-like environment with agent switching, macros, and persistent team context so you can run, orchestrate, and iterate on multi-agent systems directly from the terminal. It supports both API-hosted models (via a lightweight “lite” install) and local inference (Diffusers/Transformers/Torch stack), plus voice, vision, image and video generation commands (/yap, /ots, /vixynt, /roll), making it useful across edge, local and cloud scenarios. Technically, npcsh favors declarative, data-driven composition: NPC profiles (.npc), Jinja execution templates (.jinx) and team contexts (.ctx) are plain YAML, so personas, prompt engineering and orchestration logic are versionable and reusable. Integration with LiteLLM and providers like OpenAI, Anthropic, Gemini, Ollama, Deepseek, Mistral, and local runtimes lets teams run RAG workflows (/rag), maintain chat history (npcpy), spin up MCP orchestration (/corca), and serve teams (/serve). For practitioners this means rapid prototyping of multi-agent pipelines, reproducible context/state management between sessions, and a low-friction path to scale agent teams without deeply embedding orchestration into code.