charmbracelet/crush

Crush

Your new coding bestie, now available in your favourite terminal.
Your tools, your code, and your workflows, wired into your LLM of choice.

终端里的编程新搭档，
无缝接入你的工具、代码与工作流，全面兼容主流 LLM 模型。

Features

• Multi-Model: choose from a wide range of LLMs or add your own via OpenAI- or Anthropic-compatible APIs
• Flexible: switch LLMs mid-session while preserving context
• Session-Based: maintain multiple work sessions and contexts per project
• LSP-Enhanced: Crush uses LSPs for additional context, just like you do
• Extensible: add capabilities via MCPs (http, stdio, and sse)
• Works Everywhere: first-class support in every terminal on macOS, Linux, Windows (PowerShell and WSL), Android, FreeBSD, OpenBSD, and NetBSD
• Industrial Grade: built on the Charm ecosystem, powering 25k+ applications, from leading open source projects to business-critical infrastructure

Installation

Use a package manager:

# Homebrew
brew install charmbracelet/tap/crush
NPM
npm install -g @charmland/crush
Arch Linux (btw)
yay -S crush-bin
Nix
nix run github:numtide/nix-ai-tools#crush
FreeBSD
pkg install crush

Windows users:

# Winget
winget install charmbracelet.crush
Scoop
scoop bucket add charm https://github.com/charmbracelet/scoop-bucket.git
scoop install crush

# Add the NUR channel.
nix-channel --add https://github.com/nix-community/NUR/archive/main.tar.gz nur
nix-channel --update
Get Crush in a Nix shell.
nix-shell -p '(import  { pkgs = import  {}; }).repos.charmbracelet.crush'

{
  inputs = {
    nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
    nur.url = "github:nix-community/NUR";
  };
outputs = { self, nixpkgs, nur, ... }: {
    nixosConfigurations.your-hostname = nixpkgs.lib.nixosSystem {
      system = "x86_64-linux";
      modules = [
        nur.modules.nixos.default
        nur.repos.charmbracelet.modules.crush
        {
          programs.crush = {
            enable = true;
            settings = {
              providers = {
                openai = {
                  id = "openai";
                  name = "OpenAI";
                  base_url = "https://api.openai.com/v1";
                  type = "openai";
                  api_key = "sk-fake123456789abcdef...";
                  models = [
                    {
                      id = "gpt-4";
                      name = "GPT-4";
                    }
                  ];
                };
              };
              lsp = {
                go = { command = "gopls"; enabled = true; };
                nix = { command = "nil"; enabled = true; };
              };
              options = {
                context_paths = [ "/etc/nixos/configuration.nix" ];
                tui = { compact_mode = true; };
                debug = false;
              };
            };
          };
        }
      ];
    };
  };
}

sudo mkdir -p /etc/apt/keyrings
curl -fsSL https://repo.charm.sh/apt/gpg.key | sudo gpg --dearmor -o /etc/apt/keyrings/charm.gpg
echo "deb [signed-by=/etc/apt/keyrings/charm.gpg] https://repo.charm.sh/apt/  " | sudo tee /etc/apt/sources.list.d/charm.list
sudo apt update && sudo apt install crush

echo '[charm]
name=Charm
baseurl=https://repo.charm.sh/yum/
enabled=1
gpgcheck=1
gpgkey=https://repo.charm.sh/yum/gpg.key' | sudo tee /etc/yum.repos.d/charm.repo
sudo yum install crush

Or, download it:

• [Packages][releases] are available in Debian and RPM formats
• [Binaries][releases] are available for Linux, macOS, Windows, FreeBSD, OpenBSD, and NetBSD

Or just install it with Go:

go install github.com/charmbracelet/crush@latest

On illumos (OpenIndiana, OmniOS), the command above works as-is. Only native OS notifications are unavailable there; terminal-based notifications (OSC) and the terminal bell still work. On Oracle Solaris, add -tags sqlite3_dotlk so the local database uses dot-file locking:

go install -tags sqlite3_dotlk github.com/charmbracelet/crush@latest

[!WARNING]

Productivity may increase when using Crush and you may find yourself nerd

sniped when first using the application. If the symptoms persist, join the

[Slack][slack] or [Discord][discord] and nerd snipe the rest of us.

Getting Started

The quickest way to get started is to choose a [Hyper][hyper] model from model picker. Follow the steps to authenticate and you'll be good to go.

[Hyper], from Charm, is the official Crush provider. It’s subscription-based, with a free tier, and optimized for Crush. It’s privacy focused, with zero data retention (ZDR) is and designed to comply with GDPR. [More on Hyper][hyper].

API Keys

You can also use Crush with many other providers such as Anthopic, OpenAI, Gemini, OpenRouter and so on. Press ctrl+l to open the model picker, choose the provider of your choice, and paste your API key.

That said, you can also set environment variables for preferred providers:

| Environment Variable | Provider | | --------------------------- | -------------------------------------------------- | | HYPER_API_KEY | [Charm Hyper][hyper] | | ANTHROPIC_API_KEY | Anthropic | | OPENAI_API_KEY | OpenAI | | VERCEL_API_KEY | Vercel AI Gateway | | GEMINI_API_KEY | Google Gemini | | ZAI_API_KEY | Z.ai | | MINIMAX_API_KEY | MiniMax | | SYNTHETIC_API_KEY | Synthetic | | HF_TOKEN | Hugging Face Inference | | CEREBRAS_API_KEY | Cerebras | | OPENROUTER_API_KEY | OpenRouter | | IONET_API_KEY | io.net | | ALIBABA_SINGAPORE_API_KEY | Alibaba (Singapore) | | ALIBABA_US_API_KEY | Alibaba (United States) | | GROQ_API_KEY | Groq | | AVIAN_API_KEY | Avian | | OPENCODE_API_KEY | OpenCode Zen & Go | | VERTEXAI_PROJECT | Google Cloud VertexAI (Gemini) | | VERTEXAI_LOCATION | Google Cloud VertexAI (Gemini) | | AWS_ACCESS_KEY_ID | Amazon Bedrock (Claude) | | AWS_SECRET_ACCESS_KEY | Amazon Bedrock (Claude) | | AWS_REGION | Amazon Bedrock (Claude) | | AWS_PROFILE | Amazon Bedrock (Custom Profile) | | AWS_BEARER_TOKEN_BEDROCK | Amazon Bedrock | | AZURE_OPENAI_API_ENDPOINT | Azure OpenAI models | | AZURE_OPENAI_API_KEY | Azure OpenAI models (optional when using Entra ID) | | AZURE_OPENAI_API_VERSION | Azure OpenAI models | | MOONSHOT_API_KEY | Moonshot |

[hyper]: https://hyper.charm.land

Also note that Crush can support nearly any provider, including Local Models. For more info see Custom Providers below.

By the Way

Is there a provider you’d like to see in Crush? Is there an existing model that needs an update?

Crush’s default model listing is managed in Catwalk, a community-supported, open source repository of Crush-compatible models, and you’re welcome to contribute.

Configuration

[!TIP]

Crush ships with a builtin skill for configuring itself. Most of the time

you can just tell what you want it to configure and it will get the job done.

Crush runs great with no configuration. That said, if you do need or want to customize Crush, you can, with a crushrc.

A crushrc is just Bash with some Crush-specific builtins. It’s a lot like a .bashrc, just for your Crush. Because Crush has a native, built-in Bash interpreter, Bash-based config works identically across all platforms, including Windows.

For example:

# Add Ollama.
provider add ollama --type ollama --base-url "http://localhost:11434/v1"
Register a model on Ollama.
model add ollama/llama3.3 --name "Llama 3.3" --context-window 128000
Auto-approve some tools.
permissions allow view edit
Include some other file on a specific machine.
if [[ $HOSTNAME == "babysquid" ]]; then
    source ~/my-stuff/babysquid.sh
fi
Add an MCP server, with a GitHub API token stored in 1Password.
mcp add github \
  --type http \
  --url "https://api.github.com/mcp/" \
  --header Authorization "Bearer $(op read 'op://my-secret-key')"

Configuration can be added either local to the project itself, or globally, with the following priority:

| Priority | Unix-like | Windows | | -------- | ------------------------- | ------------------------------------- | | 1 | ./.crushrc | .\.crushrc | | 2 | ./crushrc | .\crushrc | | 3 | ~/.config/crush/crushrc | %USERPROFILE%\.config\crush\crushrc |

(Crush respects the [XDG Base Directory Specification][xdg], so your paths may differ depending on your XDG_CONFIG_HOME value. Data directories such as ~/.local/share/crush and %LOCALAPPDATA%\crush contain JSON state only; Crush does not execute a crushrc from them.)

[xdg]: https://specifications.freedesktop.org/basedir-spec/basedir-spec-latest.html

What about the old JSON format? It’s still supported, but it should be considered deprecated. See: the config docs for details.

[!TIP]

You can override the user and data config locations by setting:

> - CRUSH_GLOBAL_CONFIG

- CRUSH_GLOBAL_DATA

As an additional note, Crush also stores ephemeral data, such as application state, in one additional location. This is state and should not be edited by hand, nor should it be considered configuration.

# Unix
$HOME/.local/share/crush/crush.json
Windows
%LOCALAPPDATA%\crush\crush.json

A note on security

Both crushrc and crush.json are trusted code; crushrc runs in a full shell, and any $(...) in crush.json runs at load time. Don't launch Crush in a directory whose config you haven't reviewed, and don't randomly source files from the internet into your config.

Environment Variables

The top-level env field sets environment variables at startup, before providers are configured. This is useful for variables that affect provider authentication (e.g. the AWS SDK credential chain) without wrapping the crush command in a shell script or exporting them in your shell profile:

{
  "$schema": "https://charm.land/crush.json",
  "env": {
    "AWS_PROFILE": "my-sso-profile"
  }
}

Values support the same $VAR and $(command) expansion as other config fields, so you can reference existing environment variables or shell out for a value.

Themes

Crush ships with built-in color themes.

Switching Themes

Open the command palette with ctrl+p, select Themes, and browse the list. The UI previews each theme as you navigate, and pressing enter confirms the selection. Press esc to cancel and revert.

Editing Themes

Open Themes, highlight the theme to customize, and press ctrl+e. Changes preview live as you type. Press enter or ctrl+s to save, or esc to cancel and revert. User themes are stored globally in the Crush config directory under themes/.

You can also select a theme directly in your config with active_theme:

{
  "$schema": "https://charm.land/crush.json",
  "options": {
    "tui": {
      "active_theme": "gruvbox-dark"
    }
  }
}

Custom theme palettes are stored as JSON files in the global theme directory. For example, ~/.config/crush/themes/my-theme.json:

{
  "base": "gruvbox-dark",
  "primary": "#ff6b6b",
  "bg_base": "#1a1a2e"
}

Select it by setting active_theme to my-theme or from the Themes dialog.

Built-In Themes

| Theme | Name | | --- | --- | | Charmtone Pantera | charmtone-panther (default) | | Gruvbox Dark | gruvbox-dark |

LSPs

Crush can use LSPs for additional context to help inform its decisions, just like you would. LSPs can be added manually like so:

# crushrc
lsp add go --command "gopls" --env "GOTOOLCHAIN go1.24.5"
lsp add typescript --command "typescript-language-server" --args --stdio
lsp add nix --command "nil"

MCPs

Crush also supports Model Context Protocol (MCP) servers through three transport types: stdio for command-line servers, http for HTTP endpoints, and sse for Server-Sent Events.

# crushrc
Add a local MCP server that runs a Node.js script.
mcp add filesystem --command node --args /path/to/mcp-server.js \
  --timeout 10 --disabled-tools some-tool-name --env NODE_ENV production
Add a GitHub MCP server that uses an API token.
mcp add github --type http --url https://api.github.com/mcp/ \
  --timeout 10 --header Authorization "Bearer $GH_PAT" \
  --disabled-tools create_issue --disabled-tools create_pull_request
Add a streaming MCP server that uses SSE.
mcp add streaming-service --type sse --url "https://example.com/mcp/sse" \
  --timeout 10 --header API-Key "$API_KEY"

MCP OAuth

HTTP and SSE MCP servers that require OAuth can use Crush's built-in authorization-code flow instead of a static Authorization header. Set "oauth": true to enable it:

{
  "mcp": {
    "linear": {
      "type": "http",
      "url": "https://mcp.linear.app/mcp",
      "oauth": true
    }
  }
}

Pre-registered clients

Some servers (GitHub, Slack) don't support dynamic client registration. For those, register an OAuth app with the provider and supply the credentials directly. All values support shell expansion:

{
  "mcp": {
    "github": {
      "type": "http",
      "url": "https://api.github.com/mcp/",
      "oauth": true,
      "oauth_client_id": "Iv1.abc123def456",
      "oauth_client_secret": "$GITHUB_MCP_SECRET",
      "oauth_callback_port": 40704
    }
  }
}

When oauth_client_id is set, Crush skips dynamic client registration and authenticates as the specified client. When omitted, Crush attempts dynamic registration automatically (works with Linear, Notion, and other servers that support RFC 7591).

Sessionless servers

Some HTTP MCP servers are sessionless — they never issue a Mcp-Session-Id and reject the subscriptions/listen stream Crush opens for list-changed notifications, which would otherwise break the connection. Crush auto-detects known sessionless servers (GitHub MCP, api.githubcopilot.com/mcp), so those need no extra configuration.

For other sessionless servers, mark them explicitly with "sessionless": true (or --sessionless true in crushrc); set it to false to force the default behavior for an auto-detected URL. The tradeoff is that a sessionless server won't push live tool/prompt/resource list-changed notifications.

Hooks

Crush has preliminary support for hooks. For details, see the hook guide.

Sharing a workspace across clients

When Crush is run against a shared backend (for example two TUIs talking to the same crush serve), clients are grouped into workspaces keyed by their resolved --cwd. Two clients with the same --cwd join the same underlying workspace, so they share the session list, message history, permission queue, LSP, and MCP state.

Joining is implicit: pointing a second client at the same working directory attaches it to the existing workspace. Each new invocation, however, starts in its own fresh session by default. To pick up the conversation another client already has open, use the session manager (the session picker) and select it. Sessions surface two signals there:

• IsBusy is set while an agent turn is in flight for that session.
• AttachedClients reports how many clients are currently viewing it.

The first client to create a workspace fixes its process-wide flags. In particular, --yolo and --debug follow a first-wins rule: later clients that arrive at the same --cwd with different values for those flags do not change the running workspace. A debug log line is emitted recording the mismatch, and the workspace keeps the flags it was created with.

A workspace lives as long as at least one client has an SSE event stream open against it. When the last stream disconnects, the workspace is torn down. There is a short grace window right after POST /v1/workspaces so a client that has created the workspace but not yet opened its event stream does not get reaped before it can attach.

Global context files

Crush automatically includes two files for cross-project instructions. Think of these are personal additions to the system prompt.

• ~/.config/crush/CRUSH.md: Crush-specific rules that would confuse other

• ~/.config/AGENTS.md: generic instructions that other coding tools might

You can customize these paths with option global-context-path. Repeat the command to add multiple paths:

# Load a single markdown file.
option global-context-path "~/path/to/custom/context/file.md"
Recursively load all Markdown files in the folder.
option global-context-path "/full/path/to/folder/of/files/"

Ignoring Files

Crush respects .gitignore files by default, but you can also create a .crushignore file to specify additional files and directories that Crush should ignore. This is useful for excluding files that you want in version control but don't want Crush to consider when providing context.

The .crushignore file uses the same syntax as .gitignore and can be placed in the root of your project or in subdirectories.

Allowing Tools

By default, Crush will ask you for permission before running tool calls. If you'd like, you can allow tools to be executed without prompting you for permissions. Use this with care.

permissions allow view ls grep edit mcp_context7_get-library-doc

Disabling Built-In Tools

You can also deny tools, hiding then from the agent entirely:

permissions deny bash sourcegraph

To disable tools from MCP servers, see the MCP config section.

You only live once

You can also skip all permission prompts completely by running Crush with the --yolo flag. Be very, very careful with this feature.

Disabling Skills

You can prevent Crush from using certain skills entirely. Disabled skills are hidden from the agent, including builtin skills and skills discovered from disk.

option disable-skill crush-config

Agent Skills

Crush supports the Agent Skills open standard for extending agent capabilities with reusable skill packages. Skills are folders containing a SKILL.md file with instructions that Crush can discover and activate on demand.

The global paths we looks for skills are:

• $CRUSH_SKILLS_DIR
• $XDG_CONFIG_HOME/agents/skills or ~/.config/agents/skills/
• $XDG_CONFIG_HOME/crush/skills or ~/.config/crush/skills/
• ~/.agents/skills/
• ~/.claude/skills/
• On Windows, we _also_ look at
• %LOCALAPPDATA%\agents\skills\ or %USERPROFILE%\AppData\Local\agents\skills\
• %LOCALAPPDATA%\crush\skills\ or %USERPROFILE%\AppData\Local\crush\skills\
• Additional paths configured via options.skills_paths

• .agents/skills
• .crush/skills
• .claude/skills
• .cursor/skills

option skill-path "$HOME/squid-skills" "./other-skills"

You can get started with example skills from anthropics/skills:

# Unix
mkdir -p ~/.config/crush/skills
cd ~/.config/crush/skills
git clone https://github.com/anthropics/skills.git _temp
mv _temp/skills/* . && rm -rf _temp

# Windows (PowerShell)
mkdir -Force "$env:LOCALAPPDATA\crush\skills"
cd "$env:LOCALAPPDATA\crush\skills"
git clone https://github.com/anthropics/skills.git _temp
mv _temp/skills/* . ; rm -r -force _temp

User-Invocable Skills

Skills can be made invocable as commands from the commands palette (ctrl+p). Add user-invocable: true to the skill's YAML frontmatter:

---
name: my-hot-skill
description: A skill that can be invoked as a command.
user-invocable: true
---

User-invocable skills appear in the commands palette with a user: or project: prefix:

• Skills from global directories show as user:skill-name
• Skills from project directories show as project:skill-name

To prevent the model from auto-triggering a skill (while still allowing user invocation), add disable-model-invocation: true:

```yaml --- name: my-skill description: Only invocable by us