{"id":29936311,"url":"https://github.com/charmbracelet/crush","last_synced_at":"2026-10-27T20:19:50.022Z","repository":{"id":307157953,"uuid":"987670088","full_name":"charmbracelet/crush","owner":"charmbracelet","description":"Glamourous agentic coding for all 💘","archived":false,"fork":false,"pushed_at":"2026-10-21T15:01:14.000Z","size":32747,"stargazers_count":28734,"open_issues_count":549,"forks_count":2320,"subscribers_count":140,"default_branch":"main","last_synced_at":"2026-10-21T17:22:55.424Z","etag":null,"topics":["agentic-ai","ai","llms","ravishing"],"latest_commit_sha":null,"homepage":"","language":"Go","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"other","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/charmbracelet.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE.md","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":"AGENTS.md","claude":null,"gemini":null,"cursor":null,"copilot":null,"dco":null,"cla":"CLA.md","disclosure":null}},"created_at":"2025-05-21T12:14:57.000Z","updated_at":"2026-10-21T16:52:56.000Z","dependencies_parsed_at":"2026-10-07T16:11:05.373Z","dependency_job_id":"a58d9fea-0a87-415d-bb7d-f351f6fea5de","html_url":"https://github.com/charmbracelet/crush","commit_stats":null,"previous_names":["charmbracelet/crush","charmbracelet/opencode"],"tags_count":210,"template":false,"template_full_name":null,"purl":"pkg:github/charmbracelet/crush","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/charmbracelet%2Fcrush","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/charmbracelet%2Fcrush/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/charmbracelet%2Fcrush/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/charmbracelet%2Fcrush/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/charmbracelet","download_url":"https://codeload.github.com/charmbracelet/crush/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/charmbracelet%2Fcrush/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":344956071,"owners_count":38443830,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-10-03T21:59:58.778Z","status":"online","status_checked_at":"2026-10-21T02:00:07.576Z","response_time":57,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agentic-ai","ai","llms","ravishing"],"created_at":"2025-08-02T22:02:04.014Z","updated_at":"2026-10-22T20:14:04.101Z","avatar_url":"https://github.com/charmbracelet.png","language":"Go","funding_links":[],"categories":["Go","HarmonyOS","\u003ca name=\"programming\"\u003e\u003c/a\u003eProgramming","AI","💻 Terminal-Based AI Agents","Tools and Code","Coding agent products (IDEs, CLIs, full suites)","Agent workflow highlights","Other","Table of Contents","Tools","Agent Harness 🚀","CLI \u0026 Terminal Tools","命令行工具","CLI","Available on GitHub","Agents \u0026 Orchestration","ai","Alternatives to Claude Code","1. Local Agents","Platform Engineering","💻 コーディングエージェント","AI 机器学习","💻 Terminal \u0026 CLI Agents","Terminal-Based AI Agents","📚 Tools","AI Agent Frameworks","AI Coding Assistants","Agent SDK / framework","Terminal-native coding agents","Integrations","Catalog","🔭 Tools","⌨️ CLI Coding Tools","Agentic AI ![](https://img.shields.io/badge/_-AGENTIC-22d3ee?style=flat-square\u0026logo=openai\u0026logoColor=white)","AI Coding Assistants \u0026 Agents","Skills (144 total)","📊 The List","Runtime","Ranked AI Coding Agents"],"sub_categories":["Windows Manager","Vibe Coding and AI Coding Assistants","Or hire a skeleton","General Purpose Libraries","Coding Agents","5. The Convergence: Harness-as-a-Service (HaaS)","AI Coding CLIs","Coding Agents \u0026 IDE Extensions","AI Coding Tools","ターミナル / CLI エージェント","Agent","🌟 What You'll Find Here:","CLI Tools","Open Source","Reference Harness Implementations","Open-source multi-provider agents","Coding agents and AI IDEs","Interactive Pair Programmers","Coding","AI Terminal"],"readme":"# Crush\n\n\u003cp align=\"center\"\u003e\n    \u003ca href=\"https://stuff.charm.sh/crush/charm-crush.png\"\u003e\u003cimg width=\"450\" alt=\"Charm Crush Logo\" src=\"https://github.com/user-attachments/assets/cf8ca3ce-8b02-43f0-9d0f-5a331488da4b\" /\u003e\u003c/a\u003e\u003cbr /\u003e\n    \u003ca href=\"https://github.com/charmbracelet/crush/releases\"\u003e\u003cimg src=\"https://img.shields.io/github/release/charmbracelet/crush\" alt=\"Latest Release\"\u003e\u003c/a\u003e\n    \u003ca href=\"https://github.com/charmbracelet/crush/actions\"\u003e\u003cimg src=\"https://github.com/charmbracelet/crush/actions/workflows/build.yml/badge.svg\" alt=\"Build Status\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003eYour new coding bestie, now available in your favourite terminal.\u003cbr /\u003eYour tools, your code, and your workflows, wired into your LLM of choice.\u003c/p\u003e\n\u003cp align=\"center\"\u003e终端里的编程新搭档，\u003cbr /\u003e无缝接入你的工具、代码与工作流，全面兼容主流 LLM 模型。\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\u003cimg width=\"800\" alt=\"Crush Demo\" src=\"https://github.com/user-attachments/assets/58280caf-851b-470a-b6f7-d5c4ea8a1968\" /\u003e\u003c/p\u003e\n\n## Features\n\n- **Multi-Model:** choose from a wide range of LLMs or add your own via OpenAI- or Anthropic-compatible APIs\n- **Flexible:** switch LLMs mid-session while preserving context\n- **Session-Based:** maintain multiple work sessions and contexts per project\n- **LSP-Enhanced:** Crush uses LSPs for additional context, just like you do\n- **Extensible:** add capabilities via MCPs (`http`, `stdio`, and `sse`)\n- **Works Everywhere:** first-class support in every terminal on macOS, Linux, Windows (PowerShell and WSL), Android, FreeBSD, OpenBSD, and NetBSD\n- **Industrial Grade:** built on the Charm ecosystem, powering 25k+ applications, from leading open source projects to business-critical infrastructure\n\n## Installation\n\nUse a package manager:\n\n```bash\n# Homebrew\nbrew install charmbracelet/tap/crush\n\n# NPM\nnpm install -g @charmland/crush\n\n# Arch Linux (btw)\nyay -S crush-bin\n\n# Nix\nnix run github:numtide/nix-ai-tools#crush\n\n# FreeBSD\npkg install crush\n```\n\nWindows users:\n\n```bash\n# Winget\nwinget install charmbracelet.crush\n\n# Scoop\nscoop bucket add charm https://github.com/charmbracelet/scoop-bucket.git\nscoop install crush\n```\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003eNix (NUR)\u003c/strong\u003e\u003c/summary\u003e\n\nCrush is available via the official Charm [NUR](https://github.com/nix-community/NUR) in `nur.repos.charmbracelet.crush`, which is the most up-to-date way to get Crush in Nix.\n\nYou can also try out Crush via the NUR with `nix-shell`:\n\n```bash\n# Add the NUR channel.\nnix-channel --add https://github.com/nix-community/NUR/archive/main.tar.gz nur\nnix-channel --update\n\n# Get Crush in a Nix shell.\nnix-shell -p '(import \u003cnur\u003e { pkgs = import \u003cnixpkgs\u003e {}; }).repos.charmbracelet.crush'\n```\n\n### NixOS \u0026 Home Manager Module Usage via NUR\n\nCrush provides NixOS and Home Manager modules via NUR.\nYou can use these modules directly in your flake by importing them from NUR. Since it auto detects whether it's a home manager or nixos context you can use the import the exact same way :)\n\n```nix\n{\n  inputs = {\n    nixpkgs.url = \"github:NixOS/nixpkgs/nixos-unstable\";\n    nur.url = \"github:nix-community/NUR\";\n  };\n\n  outputs = { self, nixpkgs, nur, ... }: {\n    nixosConfigurations.your-hostname = nixpkgs.lib.nixosSystem {\n      system = \"x86_64-linux\";\n      modules = [\n        nur.modules.nixos.default\n        nur.repos.charmbracelet.modules.crush\n        {\n          programs.crush = {\n            enable = true;\n            settings = {\n              providers = {\n                openai = {\n                  id = \"openai\";\n                  name = \"OpenAI\";\n                  base_url = \"https://api.openai.com/v1\";\n                  type = \"openai\";\n                  api_key = \"sk-fake123456789abcdef...\";\n                  models = [\n                    {\n                      id = \"gpt-4\";\n                      name = \"GPT-4\";\n                    }\n                  ];\n                };\n              };\n              lsp = {\n                go = { command = \"gopls\"; enabled = true; };\n                nix = { command = \"nil\"; enabled = true; };\n              };\n              options = {\n                context_paths = [ \"/etc/nixos/configuration.nix\" ];\n                tui = { compact_mode = true; };\n                debug = false;\n              };\n            };\n          };\n        }\n      ];\n    };\n  };\n}\n```\n\n\u003c/details\u003e\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003eDebian/Ubuntu\u003c/strong\u003e\u003c/summary\u003e\n\n```bash\nsudo mkdir -p /etc/apt/keyrings\ncurl -fsSL https://repo.charm.sh/apt/gpg.key | sudo gpg --dearmor -o /etc/apt/keyrings/charm.gpg\necho \"deb [signed-by=/etc/apt/keyrings/charm.gpg] https://repo.charm.sh/apt/ * *\" | sudo tee /etc/apt/sources.list.d/charm.list\nsudo apt update \u0026\u0026 sudo apt install crush\n```\n\n\u003c/details\u003e\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003eFedora/RHEL\u003c/strong\u003e\u003c/summary\u003e\n\n```bash\necho '[charm]\nname=Charm\nbaseurl=https://repo.charm.sh/yum/\nenabled=1\ngpgcheck=1\ngpgkey=https://repo.charm.sh/yum/gpg.key' | sudo tee /etc/yum.repos.d/charm.repo\nsudo yum install crush\n```\n\n\u003c/details\u003e\n\nOr, download it:\n\n- [Packages][releases] are available in Debian and RPM formats\n- [Binaries][releases] are available for Linux, macOS, Windows, FreeBSD, OpenBSD, and NetBSD\n\n[releases]: https://github.com/charmbracelet/crush/releases\n\nOr just install it with Go:\n\n```\ngo install github.com/charmbracelet/crush@latest\n```\n\nOn illumos (OpenIndiana, OmniOS), the command above works as-is. Only native\nOS notifications are unavailable there; terminal-based notifications (OSC) and\nthe terminal bell still work. On Oracle Solaris, add `-tags sqlite3_dotlk` so\nthe local database uses dot-file locking:\n\n```\ngo install -tags sqlite3_dotlk github.com/charmbracelet/crush@latest\n```\n\n\u003e [!WARNING]\n\u003e Productivity may increase when using Crush and you may find yourself nerd\n\u003e sniped when first using the application. If the symptoms persist, join the\n\u003e [Slack][slack] or [Discord][discord] and nerd snipe the rest of us.\n\n## Getting Started\n\nThe quickest way to get started is to choose a [Hyper][hyper] model from model\npicker. Follow the steps to authenticate and you'll be good to go.\n\n[Hyper], from Charm, is the official Crush provider. It’s subscription-based,\nwith a free tier, and optimized for Crush. It’s privacy focused, with zero data\nretention (ZDR) is and designed to comply with GDPR. [More on Hyper][hyper].\n\n\u003cp\u003e\u003ca href=\"https://hyper.charm.land\"\u003e\u003cimg width=\"340\" height=\"200\" alt=\"Charm Hyper\" src=\"https://github.com/user-attachments/assets/50875289-7992-454d-9f14-9f790413fb5e\" /\u003e\u003c/a\u003e\u003c/p\u003e\n\n## API Keys\n\nYou can also use Crush with many other providers such as Anthopic, OpenAI,\nGemini, OpenRouter and so on. Press \u003ckbd\u003ectrl+l\u003c/kbd\u003e to open the model picker,\nchoose the provider of your choice, and paste your API key.\n\nThat said, you can also set environment variables for preferred providers:\n\n| Environment Variable        | Provider                                           |\n| --------------------------- | -------------------------------------------------- |\n| `HYPER_API_KEY`             | [Charm Hyper][hyper]                               |\n| `ANTHROPIC_API_KEY`         | Anthropic                                          |\n| `OPENAI_API_KEY`            | OpenAI                                             |\n| `VERCEL_API_KEY`            | Vercel AI Gateway                                  |\n| `GEMINI_API_KEY`            | Google Gemini                                      |\n| `ZAI_API_KEY`               | Z.ai                                               |\n| `MINIMAX_API_KEY`           | MiniMax                                            |\n| `SYNTHETIC_API_KEY`         | Synthetic                                          |\n| `HF_TOKEN`                  | Hugging Face Inference                             |\n| `CEREBRAS_API_KEY`          | Cerebras                                           |\n| `OPENROUTER_API_KEY`        | OpenRouter                                         |\n| `IONET_API_KEY`             | io.net                                             |\n| `ALIBABA_SINGAPORE_API_KEY` | Alibaba (Singapore)                                |\n| `ALIBABA_US_API_KEY`        | Alibaba (United States)                            |\n| `GROQ_API_KEY`              | Groq                                               |\n| `AVIAN_API_KEY`             | Avian                                              |\n| `OPENCODE_API_KEY`          | OpenCode Zen \u0026 Go                                  |\n| `VERTEXAI_PROJECT`          | Google Cloud VertexAI (Gemini)                     |\n| `VERTEXAI_LOCATION`         | Google Cloud VertexAI (Gemini)                     |\n| `AWS_ACCESS_KEY_ID`         | Amazon Bedrock (Claude)                            |\n| `AWS_SECRET_ACCESS_KEY`     | Amazon Bedrock (Claude)                            |\n| `AWS_REGION`                | Amazon Bedrock (Claude)                            |\n| `AWS_PROFILE`               | Amazon Bedrock (Custom Profile)                    |\n| `AWS_BEARER_TOKEN_BEDROCK`  | Amazon Bedrock                                     |\n| `AZURE_OPENAI_API_ENDPOINT` | Azure OpenAI models                                |\n| `AZURE_OPENAI_API_KEY`      | Azure OpenAI models (optional when using Entra ID) |\n| `AZURE_OPENAI_API_VERSION`  | Azure OpenAI models                                |\n| `MOONSHOT_API_KEY`          | Moonshot                                           |\n\n[hyper]: https://hyper.charm.land\n\nAlso note that Crush can support nearly any provider, including\n[Local Models](#local-models). For more info see\n[Custom Providers](#custom-providers) below.\n\n### By the Way\n\nIs there a provider you’d like to see in Crush? Is there an existing model that needs an update?\n\nCrush’s default model listing is managed in [Catwalk](https://github.com/charmbracelet/catwalk), a community-supported, open source repository of Crush-compatible models, and you’re welcome to contribute.\n\n\u003ca href=\"https://github.com/charmbracelet/catwalk\"\u003e\u003cimg width=\"174\" height=\"174\" alt=\"Catwalk Badge\" src=\"https://github.com/user-attachments/assets/95b49515-fe82-4409-b10d-5beb0873787d\" /\u003e\u003c/a\u003e\n\n## Configuration\n\n\u003e [!TIP]\n\u003e Crush ships with a builtin skill for configuring itself. Most of the time\n\u003e you can just tell what you want it to configure and it will get the job done.\n\nCrush runs great with no configuration. That said, if you do need or want to\ncustomize Crush, you can, with a `crushrc`.\n\nA `crushrc` is just Bash with some Crush-specific builtins. It’s a lot like\na `.bashrc`, just for your Crush. Because Crush has a native, built-in Bash\ninterpreter, Bash-based config works identically across all platforms, including\nWindows.\n\nFor example:\n\n```bash\n# Add Ollama.\nprovider add ollama --type ollama --base-url \"http://localhost:11434/v1\"\n\n# Register a model on Ollama.\nmodel add ollama/llama3.3 --name \"Llama 3.3\" --context-window 128000\n\n# Auto-approve some tools.\npermissions allow view edit\n\n# Include some other file on a specific machine.\nif [[ $HOSTNAME == \"babysquid\" ]]; then\n    source ~/my-stuff/babysquid.sh\nfi\n\n# Add an MCP server, with a GitHub API token stored in 1Password.\nmcp add github \\\n  --type http \\\n  --url \"https://api.github.com/mcp/\" \\\n  --header Authorization \"Bearer $(op read 'op://my-secret-key')\"\n```\n\nConfiguration can be added either local to the project itself, or globally,\nwith the following priority:\n\n| Priority | Unix-like                 | Windows                               |\n| -------- | ------------------------- | ------------------------------------- |\n| 1        | `./.crushrc`              | `.\\.crushrc`                          |\n| 2        | `./crushrc`               | `.\\crushrc`                           |\n| 3        | `~/.config/crush/crushrc` | `%USERPROFILE%\\.config\\crush\\crushrc` |\n\n(Crush respects the [XDG Base Directory Specification][xdg], so your paths\nmay differ depending on your `XDG_CONFIG_HOME` value. Data directories such as\n`~/.local/share/crush` and `%LOCALAPPDATA%\\crush` contain JSON state only; Crush\ndoes not execute a `crushrc` from them.)\n\n[xdg]: https://specifications.freedesktop.org/basedir-spec/basedir-spec-latest.html\n\nWhat about the old JSON format? It’s still supported, but it should be\nconsidered deprecated. See: [the config docs](./docs/config/) for details.\n\n\u003e [!TIP]\n\u003e You can override the user and data config locations by setting:\n\u003e\n\u003e - `CRUSH_GLOBAL_CONFIG`\n\u003e - `CRUSH_GLOBAL_DATA`\n\nAs an additional note, Crush also stores ephemeral data, such as application\nstate, in one additional location. This is state and should not be edited by\nhand, nor should it be considered configuration.\n\n```bash\n# Unix\n$HOME/.local/share/crush/crush.json\n\n# Windows\n%LOCALAPPDATA%\\crush\\crush.json\n```\n\n#### A note on security\n\nBoth `crushrc` and `crush.json` are trusted code; `crushrc` runs in a full\nshell, and any `$(...)` in `crush.json` runs at load time. Don't launch Crush\nin a directory whose config you haven't reviewed, and don't randomly `source`\nfiles from the internet into your config.\n\n### Environment Variables\n\nThe top-level `env` field sets environment variables at startup, before\nproviders are configured. This is useful for variables that affect provider\nauthentication (e.g. the AWS SDK credential chain) without wrapping the\n`crush` command in a shell script or exporting them in your shell profile:\n\n```json\n{\n  \"$schema\": \"https://charm.land/crush.json\",\n  \"env\": {\n    \"AWS_PROFILE\": \"my-sso-profile\"\n  }\n}\n```\n\nValues support the same `$VAR` and `$(command)` expansion as other config\nfields, so you can reference existing environment variables or shell out for\na value.\n\n### Themes\n\nCrush ships with built-in color themes.\n\n#### Switching Themes\n\nOpen the command palette with `ctrl+p`, select **Themes**, and browse the\nlist. The UI previews each theme as you navigate, and pressing `enter`\nconfirms the selection. Press `esc` to cancel and revert.\n\n#### Editing Themes\n\nOpen **Themes**, highlight the theme to customize, and press `ctrl+e`.\nChanges preview live as you type. Press `enter` or `ctrl+s` to save, or `esc`\nto cancel and revert. User themes are stored globally in the Crush config\ndirectory under `themes/`.\n\nYou can also select a theme directly in your config with `active_theme`:\n\n```json\n{\n  \"$schema\": \"https://charm.land/crush.json\",\n  \"options\": {\n    \"tui\": {\n      \"active_theme\": \"gruvbox-dark\"\n    }\n  }\n}\n```\n\nCustom theme palettes are stored as JSON files in the global theme directory.\nFor example, `~/.config/crush/themes/my-theme.json`:\n\n```json\n{\n  \"base\": \"gruvbox-dark\",\n  \"primary\": \"#ff6b6b\",\n  \"bg_base\": \"#1a1a2e\"\n}\n```\n\nSelect it by setting `active_theme` to `my-theme` or from the **Themes**\ndialog.\n\n#### Built-In Themes\n\n| Theme | Name |\n| --- | --- |\n| Charmtone Pantera | `charmtone-panther` (default) |\n| Gruvbox Dark | `gruvbox-dark` |\n\n### LSPs\n\nCrush can use LSPs for additional context to help inform its decisions, just\nlike you would. LSPs can be added manually like so:\n\n```bash\n# crushrc\n\nlsp add go --command \"gopls\" --env \"GOTOOLCHAIN go1.24.5\"\nlsp add typescript --command \"typescript-language-server\" --args --stdio\nlsp add nix --command \"nil\"\n```\n\n### MCPs\n\nCrush also supports Model Context Protocol (MCP) servers through three transport\ntypes: `stdio` for command-line servers, `http` for HTTP endpoints, and `sse`\nfor Server-Sent Events.\n\n```bash\n# crushrc\n\n# Add a local MCP server that runs a Node.js script.\nmcp add filesystem --command node --args /path/to/mcp-server.js \\\n  --timeout 10 --disabled-tools some-tool-name --env NODE_ENV production\n\n# Add a GitHub MCP server that uses an API token.\nmcp add github --type http --url https://api.github.com/mcp/ \\\n  --timeout 10 --header Authorization \"Bearer $GH_PAT\" \\\n  --disabled-tools create_issue --disabled-tools create_pull_request\n\n# Add a streaming MCP server that uses SSE.\nmcp add streaming-service --type sse --url \"https://example.com/mcp/sse\" \\\n  --timeout 10 --header API-Key \"$API_KEY\"\n```\n\n#### MCP OAuth\n\nHTTP and SSE MCP servers that require OAuth can use Crush's built-in\nauthorization-code flow instead of a static `Authorization` header. Set\n`\"oauth\": true` to enable it:\n\n```json\n{\n  \"mcp\": {\n    \"linear\": {\n      \"type\": \"http\",\n      \"url\": \"https://mcp.linear.app/mcp\",\n      \"oauth\": true\n    }\n  }\n}\n```\n\n##### Pre-registered clients\n\nSome servers (GitHub, Slack) don't support dynamic client registration.\nFor those, register an OAuth app with the provider and supply the\ncredentials directly. All values support shell expansion:\n\n```json\n{\n  \"mcp\": {\n    \"github\": {\n      \"type\": \"http\",\n      \"url\": \"https://api.github.com/mcp/\",\n      \"oauth\": true,\n      \"oauth_client_id\": \"Iv1.abc123def456\",\n      \"oauth_client_secret\": \"$GITHUB_MCP_SECRET\",\n      \"oauth_callback_port\": 40704\n    }\n  }\n}\n```\n\nWhen `oauth_client_id` is set, Crush skips dynamic client registration\nand authenticates as the specified client. When omitted, Crush attempts\ndynamic registration automatically (works with Linear, Notion, and other\nservers that support RFC 7591).\n\n#### Sessionless servers\n\nSome HTTP MCP servers are sessionless — they never issue a\n`Mcp-Session-Id` and reject the `subscriptions/listen` stream Crush opens\nfor list-changed notifications, which would otherwise break the\nconnection. Crush auto-detects known sessionless servers (GitHub MCP,\n`api.githubcopilot.com/mcp`), so those need no extra configuration.\n\nFor other sessionless servers, mark them explicitly with\n`\"sessionless\": true` (or `--sessionless true` in `crushrc`); set it to\n`false` to force the default behavior for an auto-detected URL. The\ntradeoff is that a sessionless server won't push live\ntool/prompt/resource list-changed notifications.\n\n### Hooks\n\nCrush has preliminary support for hooks. For details, see\n[the hook guide](./docs/hooks/).\n\n### Sharing a workspace across clients\n\nWhen Crush is run against a shared backend (for example two TUIs talking to\nthe same `crush serve`), clients are grouped into **workspaces** keyed by\ntheir resolved `--cwd`. Two clients with the same `--cwd` join the same\nunderlying workspace, so they share the session list, message history,\npermission queue, LSP, and MCP state.\n\nJoining is implicit: pointing a second client at the same working directory\nattaches it to the existing workspace. Each new invocation, however, starts\nin its own fresh session by default. To pick up the conversation another\nclient already has open, use the session manager (the session picker) and\nselect it. Sessions surface two signals there:\n\n- `IsBusy` is set while an agent turn is in flight for that session.\n- `AttachedClients` reports how many clients are currently viewing it.\n\nA non-zero `AttachedClients` (often combined with `IsBusy`) is the cue that a\nsession is \"in progress\" on another client and joining it will mirror that\nview live.\n\nThe first client to create a workspace fixes its process-wide flags. In\nparticular, `--yolo` and `--debug` follow a **first-wins** rule: later\nclients that arrive at the same `--cwd` with different values for those\nflags do not change the running workspace. A debug log line is emitted\nrecording the mismatch, and the workspace keeps the flags it was created\nwith.\n\nA workspace lives as long as at least one client has an SSE event stream\nopen against it. When the last stream disconnects, the workspace is torn\ndown. There is a short grace window right after `POST /v1/workspaces` so a\nclient that has created the workspace but not yet opened its event stream\ndoes not get reaped before it can attach.\n\n### Global context files\n\nCrush automatically includes two files for cross-project instructions. Think of\nthese are personal additions to the system prompt.\n\n- `~/.config/crush/CRUSH.md`: Crush-specific rules that would confuse other\n  agentic coding tools. If you only use Crush, this is the only one you need to\n  edit.\n- `~/.config/AGENTS.md`: generic instructions that other coding tools might\n  read. Avoid referring to Crush-specific features or workflows here. You\n  probably only care about this if you use multiple agentic coding tools and\n  want to share instructions between them.\n\nYou can customize these paths with `option global-context-path`. Repeat the\ncommand to add multiple paths:\n\n```bash\n# Load a single markdown file.\noption global-context-path \"~/path/to/custom/context/file.md\"\n\n# Recursively load all Markdown files in the folder.\noption global-context-path \"/full/path/to/folder/of/files/\"\n```\n\n### Ignoring Files\n\nCrush respects `.gitignore` files by default, but you can also create a\n`.crushignore` file to specify additional files and directories that Crush\nshould ignore. This is useful for excluding files that you want in version\ncontrol but don't want Crush to consider when providing context.\n\nThe `.crushignore` file uses the same syntax as `.gitignore` and can be placed\nin the root of your project or in subdirectories.\n\n### Allowing Tools\n\nBy default, Crush will ask you for permission before running tool calls. If\nyou'd like, you can allow tools to be executed without prompting you for\npermissions. Use this with care.\n\n```bash\npermissions allow view ls grep edit mcp_context7_get-library-doc\n```\n\n### Disabling Built-In Tools\n\nYou can also deny tools, hiding then from the agent entirely:\n\n```bash\npermissions deny bash sourcegraph\n```\n\nTo disable tools from MCP servers, see the [MCP config section](#mcps).\n\n### You only live once\n\nYou can also skip all permission prompts completely by running Crush with the\n`--yolo` flag. Be very, very careful with this feature.\n\n### Disabling Skills\n\nYou can prevent Crush from using certain skills entirely. Disabled skills are\nhidden from the agent, including builtin skills and skills discovered from\ndisk.\n\n```bash\noption disable-skill crush-config\n```\n\n### Agent Skills\n\nCrush supports the [Agent Skills](https://agentskills.io) open standard for\nextending agent capabilities with reusable skill packages. Skills are folders\ncontaining a `SKILL.md` file with instructions that Crush can discover and\nactivate on demand.\n\nThe global paths we looks for skills are:\n\n- `$CRUSH_SKILLS_DIR`\n- `$XDG_CONFIG_HOME/agents/skills` or `~/.config/agents/skills/`\n- `$XDG_CONFIG_HOME/crush/skills` or `~/.config/crush/skills/`\n- `~/.agents/skills/`\n- `~/.claude/skills/`\n- On Windows, we _also_ look at\n  - `%LOCALAPPDATA%\\agents\\skills\\` or `%USERPROFILE%\\AppData\\Local\\agents\\skills\\`\n  - `%LOCALAPPDATA%\\crush\\skills\\` or `%USERPROFILE%\\AppData\\Local\\crush\\skills\\`\n- Additional paths configured via `options.skills_paths`\n\nOn top of that, we _also_ load skills in your project from the following\nrelative paths:\n\n- `.agents/skills`\n- `.crush/skills`\n- `.claude/skills`\n- `.cursor/skills`\n\nOr load directories of skills specifically in your config:\n\n```bash\noption skill-path \"$HOME/squid-skills\" \"./other-skills\"\n```\n\nYou can get started with example skills from [anthropics/skills](https://github.com/anthropics/skills):\n\n```bash\n# Unix\nmkdir -p ~/.config/crush/skills\ncd ~/.config/crush/skills\ngit clone https://github.com/anthropics/skills.git _temp\nmv _temp/skills/* . \u0026\u0026 rm -rf _temp\n```\n\n```powershell\n# Windows (PowerShell)\nmkdir -Force \"$env:LOCALAPPDATA\\crush\\skills\"\ncd \"$env:LOCALAPPDATA\\crush\\skills\"\ngit clone https://github.com/anthropics/skills.git _temp\nmv _temp/skills/* . ; rm -r -force _temp\n```\n\n#### User-Invocable Skills\n\nSkills can be made invocable as commands from the commands palette\n(\u003ckbd\u003ectrl+p\u003c/kbd\u003e). Add `user-invocable: true` to the skill's YAML\nfrontmatter:\n\n```yaml\n---\nname: my-hot-skill\ndescription: A skill that can be invoked as a command.\nuser-invocable: true\n---\n```\n\nUser-invocable skills appear in the commands palette with a `user:` or `project:` prefix:\n\n- Skills from global directories show as `user:skill-name`\n- Skills from project directories show as `project:skill-name`\n\nWhen invoked, the skill's instructions are loaded into the conversation context.\n\nTo prevent the model from auto-triggering a skill (while still allowing user invocation), add `disable-model-invocation: true`:\n\n```yaml\n---\nname: my-skill\ndescription: Only invocable by users, not the model.\nuser-invocable: true\ndisable-model-invocation: true\n---\n```\n\nSkills with `disable-model-invocation` won't appear in the model's available skills list but can still be invoked manually by users.\n\n### Desktop notifications\n\nCrush sends desktop notifications when a tool call requires permission and when\nthe agent finishes its turn. They're only sent when the terminal window isn't\nfocused _and_ your terminal supports reporting the focus state.\n\n```bash\n# Choose auto, native, osc, bell, or disabled.\noption notifications disabled\n```\n\n`auto` uses native notifications locally and OSC notifications over SSH when\nsupported.\n\n### Initialization\n\nWhen you initialize a project, Crush analyzes your codebase and creates\na context file that helps it work more effectively in future sessions. By\ndefault, this file is named `AGENTS.md`, but you can customize the name and\nlocation with the `initialize-as` option:\n\n```bash\n# crushrc\noption initialize-as AGENTS.md\n```\n\nThis is useful if you prefer a different naming convention or want to place the\nfile in a specific directory (e.g., `CRUSH.md` or `docs/LLMs.md`). Crush will\nfill the file with project-specific context like build commands, code patterns,\nand conventions it discovered during initialization.\n\n### Attribution Settings\n\nBy default, Crush adds attribution information to Git commits and pull requests\nit creates. You can customize this behavior with `option` commands:\n\n```bash\noption attribution-trailer-style co-authored-by\noption attribution-generated-with true\n```\n\n- `trailer_style`: Controls the attribution trailer added to commit messages\n  (default: `assisted-by`)\n  - `assisted-by`: Adds `Assisted-by: Crush:[ModelID]` as specified in [the convention](https://docs.kernel.org/process/coding-assistants.html#attribution)\n  - `co-authored-by`: Adds `Co-Authored-By: Crush \u003ccrush@charm.land\u003e`\n  - `none`: No attribution trailer\n- `generated_with`: When true (default), adds `💘 Generated with Crush` line to\n  commit messages and PR descriptions\n\n### Custom Providers\n\nCrush supports custom provider configurations for both OpenAI-compatible and\nAnthropic-compatible APIs.\n\n\u003e [!NOTE]\n\u003e Note that we support two \"types\" for OpenAI. Make sure to choose the right one\n\u003e to ensure the best experience!\n\u003e\n\u003e - `openai` should be used when proxying or routing requests through OpenAI.\n\u003e - `openai-compat` should be used when using non-OpenAI providers that have OpenAI-compatible APIs.\n\n#### OpenAI-Compatible APIs\n\nHere’s an example configuration for Deepseek, which uses an OpenAI-compatible\nAPI. Don't forget to set `DEEPSEEK_API_KEY` in your environment.\n\n```bash\nprovider add deepseek --type openai-compat \\\n  --base-url \"https://api.deepseek.com/v1\" \\\n  --api-key \"$DEEPSEEK_API_KEY\"\n\nmodel add deepseek/deepseek-chat \\\n  --name \"Deepseek V3\" \\\n  --context-window 64000 \\\n  --default-max-tokens 5000 \\\n  --price-input 0.27 \\\n  --price-output 1.1 \\\n  --price-cache-create 1.1 \\\n  --price-cache-hit 0.07\n```\n\n#### Anthropic-Compatible APIs\n\nCustom Anthropic-compatible providers follow this format:\n\n```bash\nprovider add custom-anthropic \\\n  --type anthropic \\\n  --base-url \"https://api.anthropic.com/v1\" \\\n  --api-key \"$ANTHROPIC_API_KEY\" \\\n  --extra-header anthropic-version 2023-06-01\n\nmodel add custom-anthropic/claude-sonnet-4-20250514 \\\n  --name \"Claude Sonnet 4\" \\\n  --context-window 200000 \\\n  --default-max-tokens 50000 \\\n  --can-reason true \\\n  --supports-images true \\\n  --price-input 3 \\\n  --price-output 15 \\\n  --price-cache-create 3.75 \\\n  --price-cache-hit 0.3\n```\n\n### Amazon Bedrock\n\nCrush currently supports running Anthropic models through Bedrock, with caching disabled.\n\nA Bedrock provider appears once Crush can find AWS credentials. You can\nauthenticate in one of two ways:\n\n**API key.** Set `AWS_BEARER_TOKEN_BEDROCK` to a Bedrock API key. This is the\nsimplest option and never expires mid-session.\n\n**AWS credential chain (SSO, profiles, access keys).** Configure AWS the usual\nway with `aws configure` or `aws configure sso`. Crush picks up whatever the\nAWS SDK credential chain resolves, including `AWS_PROFILE`, `AWS_ACCESS_KEY_ID`\n/ `AWS_SECRET_ACCESS_KEY`, or an SSO session. To select a specific profile,\nset `AWS_PROFILE` in your shell (`AWS_PROFILE=myprofile crush`) or in the\ntop-level [`env`](#environment-variables) config.\n\nIf you authenticate via AWS SSO, your session expires periodically. Set\n`aws_auth_refresh` to a command that refreshes it. When Bedrock returns a\ncredential error, Crush runs the command, then retries the request in place\n(no duplicate messages, no manual restart):\n\n```json\n{\n  \"$schema\": \"https://charm.land/crush.json\",\n  \"env\": {\n    \"AWS_PROFILE\": \"my-sso-profile\"\n  },\n  \"providers\": {\n    \"bedrock\": {\n      \"aws_auth_refresh\": \"aws sso login --profile my-sso-profile\"\n    },\n    \"bedrock-europe\": {\n      \"aws_auth_refresh\": \"aws sso login --profile my-eu-sso-profile\"\n    }\n  }\n}\n```\n\n- `aws_auth_refresh` — shell command run when AWS credentials expire (e.g. `aws sso login`)\n\n### Vertex AI Platform\n\nVertex AI will appear in the list of available providers when `VERTEXAI_PROJECT` and `VERTEXAI_LOCATION` are set. You will also need to be authenticated:\n\n```bash\n$ gcloud auth application-default login\n```\n\nTo add specific models to the configuration, configure as such:\n\n```bash\n# crushrc — authentication still comes from gcloud and the VERTEXAI_* env vars.\nprovider add vertexai --type google-vertex\n\nmodel add vertexai/claude-sonnet-4@20250514 \\\n  --name \"VertexAI Sonnet 4\" \\\n  --context-window 200000 \\\n  --default-max-tokens 50000 \\\n  --can-reason true \\\n  --supports-images true \\\n  --price-input 3 \\\n  --price-output 15 \\\n  --price-cache-create 3.75 \\\n  --price-cache-hit 0.3\n```\n\n### Local Models\n\nCrush can auto-discovers models from local providers. Add a custom provider\nwith `type` set to `llamacpp`, `omlx`, `lmstudio`, `litellm`, or `ollama`\nand leave out the models list. Crush will populate the model list\nautomatically.\n\n```bash\n# Piece of cake.\nprovider add ollama \\\n  --name Ollama \\\n  --type ollama \\\n  --base-url \"http://localhost:11434/v1/\"\n```\n\nFor llama.cpp (`llama-server`), point at the server's base URL:\n\n```bash\nprovider add llamacpp \\\n  --name \"llama.cpp\" \\\n  --type llamacpp \\\n  --base-url \"http://localhost:2222\"\n```\n\n#### Manual Model Configuration\n\nYou can still list models explicitly. User-defined models always take\nprecedence over discovered ones, and any fields you set won't be overwritten\nby auto-discovery. Auto discovery will run if the model list is empty for any\n`openai-compat` provider or if you pass `\"discover_models\": true` it will merge\nthe found models with your hand configured ones.\n\n```bash\n# crushrc\nprovider add ollama \\\n  --name Ollama \\\n  --type ollama \\\n  --base-url \"http://localhost:11434/v1/\" \\\n  --discover-models true\n\nmodel add ollama/qwen3:30b \\\n  --name \"Qwen 3 30B\" \\\n  --context-window 256000 \\\n  --default-max-tokens 20000\n```\n\nThe `--discover-models true` flag merges discovered models with the one above;\nyour explicit model fields win on conflicts.\n\n## Logging\n\nSometimes you need to look at logs. Luckily, Crush logs all sorts of\nstuff. Logs are stored in `./.crush/logs/crush.log` relative to the project.\n\nThe CLI also contains some helper commands to make perusing recent logs easier:\n\n```bash\n# Print the last 1000 lines\ncrush logs\n\n# Print the last 500 lines\ncrush logs --tail 500\n\n# Follow logs in real time\ncrush logs --follow\n```\n\nWant more logging? Run `crush` with the `--debug` flag, or enable it in your\n`crushrc`:\n\n```bash\n# crushrc\noption debug true\noption debug-lsp true\n```\n\n## Provider Auto-Updates\n\nBy default, Crush automatically checks for the latest and greatest list of\nproviders and models from [Catwalk](https://github.com/charmbracelet/catwalk),\nthe open source Crush provider database. This means that when new providers and\nmodels are available, or when model metadata changes, Crush automatically\nupdates your local configuration.\n\n### Custom provider catalog\n\nYou can also override [Catwalk](https://github.com/charmbracelet/catwalk) default URL (for testing, using a fork).\n\nYou can do so by setting `CATWALK_URL` enviromental variable. (e.g. `export CATWALK_URL=http://localhost:8000`)\n\n### Disabling automatic provider updates\n\nFor those with restricted internet access, or those who prefer to work in\nair-gapped environments, this might not be want you want, and this feature can\nbe disabled.\n\nTo disable automatic provider updates in your `crushrc`:\n\n```bash\noption provider-auto-update false\n```\n\nOr set the `CRUSH_DISABLE_PROVIDER_AUTO_UPDATE` environment variable:\n\n```bash\nexport CRUSH_DISABLE_PROVIDER_AUTO_UPDATE=1\n```\n\n### Manually updating providers\n\nManually updating providers is possible with the `crush update-providers`\ncommand:\n\n```bash\n# Update providers remotely from Catwalk.\ncrush update-providers\n\n# Update providers from a custom Catwalk base URL.\ncrush update-providers https://example.com/\n\n# Update providers from a local file.\ncrush update-providers /path/to/local-providers.json\n\n# Reset providers to the embedded version, embedded at crush at build time.\ncrush update-providers embedded\n\n# For more info:\ncrush update-providers --help\n```\n\n## Metrics\n\nCrush records pseudonymous usage metrics (tied to a device-specific hash),\nwhich maintainers rely on to inform development and support priorities. The\nmetrics include solely usage metadata; prompts and responses are NEVER\ncollected.\n\nDetails on exactly what’s collected are in the source code ([here](https://github.com/charmbracelet/crush/tree/main/internal/event)\nand [here](https://github.com/charmbracelet/crush/blob/main/internal/llm/agent/event.go)).\n\nYou can opt out of metrics collection at any time by setting the environment\nvariable by setting the following in your environment:\n\n```bash\nexport CRUSH_DISABLE_METRICS=1\n```\n\nCrush also respects the [`DO_NOT_TRACK`](https://donottrack.sh/) convention\nwhich can be enabled via `export DO_NOT_TRACK=1`.\n\n## Q\u0026A\n\n### Why is clipboard copy and paste not working?\n\nInstalling an extra tool might be needed on Unix-like environments.\n\n| Environment         | Tool                     |\n| ------------------- | ------------------------ |\n| Windows             | Native support           |\n| macOS               | Native support           |\n| Linux/BSD + Wayland | `wl-copy` and `wl-paste` |\n| Linux/BSD + X11     | `xclip` or `xsel`        |\n\n## Contributing\n\nSee the [contributing guide](https://github.com/charmbracelet/crush?tab=contributing-ov-file#contributing).\n\n## Whatcha think?\n\nWe’d love to hear your thoughts on this project. Need help? We gotchu. You can find us on:\n\n- [Twitter](https://twitter.com/charmcli)\n- [Slack][slack]\n- [Discord][discord]\n- [The Fediverse](https://mastodon.social/@charmcli)\n- [Bluesky](https://bsky.app/profile/charm.land)\n\n[slack]: https://charm.land/slack\n[discord]: https://charm.land/discord\n\n## License\n\n[FSL-1.1-MIT](https://github.com/charmbracelet/crush/raw/main/LICENSE.md)\n\n---\n\nPart of [Charm](https://charm.land).\n\n\u003ca href=\"https://charm.land/\"\u003e\u003cimg alt=\"The Charm logo\" width=\"400\" src=\"https://stuff.charm.sh/charm-banner-softy.jpg\" /\u003e\u003c/a\u003e\n\n\u003c!--prettier-ignore--\u003e\nCharm热爱开源 • Charm loves open source\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcharmbracelet%2Fcrush","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fcharmbracelet%2Fcrush","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcharmbracelet%2Fcrush/lists"}