Crush: Terminal-based AI coding agent
LLM-powered coding agent with LSP and MCP integration.
Learn more about Crush
Crush is a terminal-based coding agent that runs on macOS, Linux, Windows, and BSD systems. It connects to language models through configurable providers and integrates with LSPs for code context, while supporting Model Context Protocol extensions via HTTP, stdio, and SSE transports. The tool maintains separate work sessions per project and allows switching between different LLM models within a session while preserving conversation history. Common deployment involves local terminal usage with remote or local LLM backends, supporting workflows that combine code editing, analysis, and generation tasks.
Multi-Provider LLM Support
Switch between OpenAI, Anthropic, and custom LLM providers mid-session without losing conversation context. Configure multiple models through simple configuration files for different tasks or cost optimization.
LSP and MCP Integration
Native Language Server Protocol integration provides deep code understanding and navigation. Model Context Protocol support with HTTP, stdio, and SSE transports enables extensibility through external tools and services.
Per-Project Session Management
Independent work sessions maintain separate conversation contexts for each project. Switch between multiple coding tasks without state collision or context loss across different codebases.
from crush import CrushConfig, Provider
config = CrushConfig()
config.add_provider(
Provider(
id="openai",
name="OpenAI",
type="openai",
base_url="https://api.openai.com/v1",
api_key="sk-your-key-here",
models=[{"id": "gpt-4o", "name": "GPT-4o"}]
)
)
config.save()Related Repositories
Discover similar tools and frameworks used by developers
vLLM
Fast, memory-efficient LLM inference engine with PagedAttention for production deployments at scale.
StabilityMatrix
Multi-backend inference UI manager with embedded dependencies.
Stable Diffusion
Text-to-image diffusion in compressed latent space.
DeepSpeed
PyTorch library for training billion-parameter models efficiently.
GroundingDINO
Zero-shot object detection from text prompts.