Crush

A terminal AI coding assistant that connects to local backends such as Ollama and llama.cpp or cloud APIs, with model switching that preserves context.

Alternative to:

Crush is a terminal AI coding assistant for developers who want an agent to work with their code and existing tools. The application runs on your machine; your chosen backend determines where model requests go.

Model choice: Connect to Ollama, llama.cpp, LM Studio, oMLX or LiteLLM, or use cloud providers such as Anthropic, OpenAI, Gemini and OpenRouter.
Session continuity: Switch models without losing conversation context and keep separate work sessions within a project.
Code context and tools: Language Server Protocol integrations supply code context, while Model Context Protocol connections give the agent access to additional tools.
Custom providers: Use OpenAI-compatible or Anthropic-compatible APIs beyond the built-in provider choices, with automatic model discovery for supported local backends.

Crush supports macOS, Linux, Windows through PowerShell and WSL, Android, FreeBSD, OpenBSD and NetBSD. Charm's Hyper is a separate hosted model provider for Crush; using it or another cloud provider sends model requests outside your local machine.

The application records pseudonymous usage metadata associated with a device-specific hash. These metrics exclude prompts and responses. You can opt out, and Crush respects DO_NOT_TRACK.