llms.txt

# OpenCode Documentation for LLMs

> OpenCode is an open-source AI coding agent available as a terminal interface, desktop application, or IDE extension. It helps developers write code, add features, and understand codebases through conversational interactions.

## Installation

Multiple installation methods are available:

```bash
# curl script (recommended)
curl -fsSL https://opencode.ai/install | bash

# Node.js package managers
npm install -g opencode
bun install -g opencode
pnpm add -g opencode
yarn global add opencode

# Homebrew (macOS/Linux)
brew install anomalyco/tap/opencode

# Arch Linux
paru -S opencode

# Windows
choco install opencode    # Chocolatey
scoop install opencode    # Scoop
# Or use Docker/WSL (recommended for Windows)
```

## Configuration

Configuration file: `opencode.json` or `opencode.jsonc` (with comments)

Schema: `https://opencode.ai/config.json`

### Core Settings

```json
{
  "$schema": "https://opencode.ai/config.json",
  "model": "anthropic/claude-sonnet-4-5",
  "small_model": "anthropic/claude-haiku-4-5",
  "default_agent": "build",
  "instructions": [
    "CONTRIBUTING.md",
    "docs/guidelines.md"
  ],
  "plugin": [
    "opencode-helicone-session",
    "./.opencode/plugins"
  ],
  "agent": { /* agent definitions */ },
  "command": { /* command definitions */ },
  "permission": {
    "edit": "ask",
    "bash": "ask",
    "mcp_*": "ask"
  },
  "tools": {
    "write": true,
    "bash": true
  }
}
```

### Environment Variables

Use `{env:VAR_NAME}` for environment variables and `{file:path}` for file contents in configuration values.

## Agents

OpenCode supports two agent types:

### Primary Agents
Main assistants you interact with directly. Switch between them using Tab or configured keybinds.

### Subagents
Specialized assistants that primary agents invoke automatically or through `@` mentions (e.g., `@general help me search`).

### Built-in Agents

| Agent | Type | Description |
|-------|------|-------------|
| build | primary | Default agent with full tool access for development work |
| plan | primary | Restricted agent for analysis; file edits and bash set to "ask" |
| general | subagent | Full tool access for multi-step research tasks |
| explore | subagent | Read-only agent for rapid codebase exploration |
| compaction | system | Hidden agent for context compaction |
| title | system | Hidden agent for title generation |
| summary | system | Hidden agent for summarization |

### Agent Configuration

JSON format in `opencode.json`:

```json
{
  "agent": {
    "code-reviewer": {
      "description": "Reviews code for best practices",
      "mode": "subagent",
      "model": "anthropic/claude-opus-4-5",
      "prompt": "{file:.opencode/prompts/agents/code-reviewer.txt}",
      "temperature": 0.3,
      "tools": {
        "write": false,
        "edit": false,
        "read": true,
        "bash": true
      },
      "permission": {
        "edit": "deny",
        "bash": {
          "*": "ask",
          "git status": "allow"
        }
      },
      "steps": 10
    }
  }
}
```

Markdown format in `~/.config/opencode/agents/` or `.opencode/agents/`:

```markdown
---
description: Expert code review specialist
mode: subagent
model: anthropic/claude-opus-4-5
temperature: 0.3
tools:
  write: false
  read: true
  bash: true
permission:
  edit: deny
steps: 10
---

You are an expert code reviewer. Review code for quality, security, and maintainability...
```

### Agent Configuration Options

| Option | Purpose | Values |
|--------|---------|--------|
| description | Required field explaining agent purpose | string |
| mode | Agent type | "primary", "subagent", or "all" |
| model | Override default model | "provider/model-id" |
| temperature | Control randomness | 0.0-1.0 (lower = focused) |
| tools | Enable/disable specific tools | object or wildcards |
| permission | Set tool permissions | "ask", "allow", or "deny" |
| steps | Limit agentic iterations | number |
| prompt | Reference custom prompt file | "{file:./path}" |
| top_p | Alternative randomness control | 0.0-1.0 |

## Commands

### Built-in Commands

- `/init` - Initialize project analysis (creates AGENTS.md)
- `/undo` - Undo last change
- `/redo` - Redo undone change
- `/share` - Generate shareable conversation link
- `/help` - Show help
- `/connect` - Configure API providers

### Custom Commands

**Markdown files** in `~/.config/opencode/commands/` (global) or `.opencode/commands/` (project):

```markdown
---
description: Create implementation plan
agent: planner
subtask: true
---

Create a detailed implementation plan for: $ARGUMENTS

Include:
- Requirements analysis
- Architecture review
- Step breakdown
- Testing strategy
```

**JSON configuration** in `opencode.json`:

```json
{
  "command": {
    "plan": {
      "description": "Create implementation plan",
      "template": "Create a detailed implementation plan for: $ARGUMENTS",
      "agent": "planner",
      "subtask": true
    },
    "test": {
      "template": "Run tests with coverage for: $ARGUMENTS\n\nOutput:\n!`npm test`",
      "description": "Run tests with coverage",
      "agent": "build"
    }
  }
}
```

### Template Variables

| Variable | Description |
|----------|-------------|
| `$ARGUMENTS` | All command arguments |
| `$1`, `$2`, `$3` | Positional arguments |
| `!`command`` | Include shell command output |
| `@filepath` | Include file contents |

### Command Options

| Option | Purpose | Required |
|--------|---------|----------|
| template | Prompt text sent to LLM | Yes |
| description | UI display text | No |
| agent | Target agent for execution | No |
| model | Override default LLM | No |
| subtask | Force subagent invocation | No |

## Tools

### Built-in Tools

| Tool | Purpose | Permission Key |
|------|---------|---------------|
| bash | Execute shell commands | "bash" |
| edit | Modify existing files using exact string replacements | "edit" |
| write | Create new files or overwrite existing ones | "edit" |
| read | Read file contents from codebase | "read" |
| grep | Search file contents using regular expressions | "grep" |
| glob | Find files by pattern matching | "glob" |
| list | List files and directories | "list" |
| lsp | Access code intelligence (experimental) | "lsp" |
| patch | Apply patches to files | "edit" |
| skill | Load skill files (SKILL.md) | "skill" |
| todowrite | Manage todo lists during sessions | "todowrite" |
| todoread | Read existing todo lists | "todoread" |
| webfetch | Fetch and read web pages | "webfetch" |
| question | Ask user questions during execution | "question" |

### Tool Permissions

```json
{
  "permission": {
    "edit": "ask",
    "bash": "allow",
    "webfetch": "deny",
    "mcp_*": "ask"
  }
}
```

Permission levels:
- `"allow"` - Tool executes without restriction
- `"deny"` - Tool cannot be used
- `"ask"` - Requires user approval before execution

## Custom Tools

Location: `.opencode/tools/` (project) or `~/.config/opencode/tools/` (global)

### Tool Definition

```typescript
import { tool } from "@opencode-ai/plugin"

export default tool({
  description: "Execute SQL query on the database",
  args: {
    query: tool.schema.string().describe("SQL query to execute"),
    database: tool.schema.string().optional().describe("Target database")
  },
  async execute(args, context) {
    // context.worktree - git repository root
    // context.directory - current working directory
    // context.sessionID - current session ID
    // context.agent - active agent identifier

    const result = await someDbQuery(args.query)
    return { result }
  }
})
```

### Multiple Tools Per File

```typescript
// math.ts - creates math_add and math_multiply tools
export const add = tool({
  description: "Add two numbers",
  args: {
    a: tool.schema.number(),
    b: tool.schema.number()
  },
  async execute({ a, b }) {
    return { result: a + b }
  }
})

export const multiply = tool({
  description: "Multiply two numbers",
  args: {
    a: tool.schema.number(),
    b: tool.schema.number()
  },
  async execute({ a, b }) {
    return { result: a * b }
  }
})
```

### Schema Types (Zod)

```typescript
tool.schema.string()
tool.schema.number()
tool.schema.boolean()
tool.schema.array(tool.schema.string())
tool.schema.object({ key: tool.schema.string() })
tool.schema.enum(["option1", "option2"])
tool.schema.optional()
tool.schema.describe("Description for LLM")
```

## Plugins

Plugins extend OpenCode with custom hooks, tools, and behaviors.

### Plugin Structure

```typescript
import { tool } from "@opencode-ai/plugin"

export const MyPlugin = async ({ project, client, $, directory, worktree }) => {
  // project - Current project information
  // client - OpenCode SDK client for AI interaction
  // $ - Bun's shell API for command execution
  // directory - Current working directory
  // worktree - Git worktree path

  return {
    // Hook implementations
    "file.edited": async (event) => {
      // Handle file edit event
    },

    "tool.execute.before": async (input, output) => {
      // Intercept before tool execution
    },

    "session.idle": async (event) => {
      // Handle session idle
    }
  }
}
```

### Loading Plugins

1. **Local files**: Place in `.opencode/plugins/` (project) or `~/.config/opencode/plugins/` (global)
2. **npm packages**: Specify in `opencode.json`:

```json
{
  "plugin": [
    "opencode-helicone-session",
    "@my-org/custom-plugin",
    "./.opencode/plugins"
  ]
}
```

### Available Hook Events

**Command Events:**
- `command.executed` - After a command is executed

**File Events:**
- `file.edited` - After a file is edited
- `file.watcher.updated` - When file watcher detects changes

**Tool Events:**
- `tool.execute.before` - Before tool execution (can modify input)
- `tool.execute.after` - After tool execution (can modify output)

**Session Events:**
- `session.created` - When session starts
- `session.compacted` - After context compaction
- `session.deleted` - When session ends
- `session.idle` - When session becomes idle
- `session.updated` - When session is updated
- `session.status` - Session status changes

**Message Events:**
- `message.updated` - When message is updated
- `message.removed` - When message is removed
- `message.part.updated` - When message part is updated

**LSP Events:**
- `lsp.client.diagnostics` - LSP diagnostic updates
- `lsp.updated` - LSP state updates

**Shell Events:**
- `shell.env` - Modify shell environment variables

**TUI Events:**
- `tui.prompt.append` - Append to TUI prompt
- `tui.command.execute` - Execute TUI command
- `tui.toast.show` - Show toast notification

**Other Events:**
- `installation.updated` - Installation updates
- `permission.asked` - Permission request
- `server.connected` - Server connection
- `todo.updated` - Todo list updates

### Hook Event Mapping (Claude Code → OpenCode)

| Claude Code Hook | OpenCode Plugin Event |
|-----------------|----------------------|
| PreToolUse | `tool.execute.before` |
| PostToolUse | `tool.execute.after` |
| Stop | `session.idle` or `session.status` |
| SessionStart | `session.created` |
| SessionEnd | `session.deleted` |
| N/A | `file.edited`, `file.watcher.updated` |
| N/A | `message.*`, `permission.*`, `lsp.*` |

### Plugin Example: Auto-Format

```typescript
export const AutoFormatPlugin = async ({ $, directory }) => {
  return {
    "file.edited": async (event) => {
      if (event.path.match(/\.(ts|tsx|js|jsx)$/)) {
        await $`prettier --write ${event.path}`
      }
    }
  }
}
```

### Plugin Example: TypeScript Check

```typescript
export const TypeCheckPlugin = async ({ $, client }) => {
  return {
    "tool.execute.after": async (input, output) => {
      if (input.tool === "edit" && input.args.filePath?.match(/\.tsx?$/)) {
        const result = await $`npx tsc --noEmit`.catch(e => e)
        if (result.exitCode !== 0) {
          client.app.log("warn", "TypeScript errors detected")
        }
      }
    }
  }
}
```

## Providers

OpenCode integrates 75+ LLM providers via AI SDK and Models.dev.

### Supported Providers

- OpenCode Zen (recommended for beginners)
- Anthropic (Claude)
- OpenAI (GPT)
- Google (Gemini)
- Amazon Bedrock
- Azure OpenAI
- GitHub Copilot
- Ollama (local)
- And 70+ more

### Provider Configuration

```json
{
  "provider": {
    "anthropic": {
      "options": {
        "baseURL": "https://api.anthropic.com/v1"
      }
    },
    "custom-provider": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Display Name",
      "options": {
        "baseURL": "https://api.example.com/v1",
        "apiKey": "{env:CUSTOM_API_KEY}"
      },
      "models": {
        "model-id": { "name": "Model Name" }
      }
    }
  }
}
```

### Model Naming Convention

Format: `provider/model-id`

Examples:
- `anthropic/claude-sonnet-4-5`
- `anthropic/claude-opus-4-5`
- `anthropic/claude-haiku-4-5`
- `openai/gpt-4o`
- `google/gemini-2.0-flash`

### API Key Setup

```bash
# Interactive setup
opencode
/connect

# Environment variables
export ANTHROPIC_API_KEY=sk-...
export OPENAI_API_KEY=sk-...
```

Keys stored in: `~/.local/share/opencode/auth.json`

## MCP (Model Context Protocol)

Configure MCP servers in `opencode.json`:

```json
{
  "mcp": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "{env:GITHUB_TOKEN}"
      }
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]
    }
  }
}
```

MCP tool permissions use `mcp_*` wildcard:

```json
{
  "permission": {
    "mcp_*": "ask"
  }
}
```

## Ecosystem Plugins

### Authentication & Provider Plugins
- Alternative model access (ChatGPT Plus, Gemini, Antigravity)
- Session tracking (Helicone headers)
- OAuth integrations

### Development Tools
- Sandbox isolation (Daytona integration)
- Type injection for TypeScript/Svelte
- DevContainer multi-branch support
- Git worktree management

### Enhancement Plugins
- Web search with citations
- Markdown table formatting
- Dynamic context token pruning
- Desktop notifications
- Persistent memory (Supermemory)
- Background process management

### Plugin Discovery

- opencode.cafe - Community plugin registry
- awesome-opencode - Curated list
- GitHub search for "opencode-plugin"

## Quick Reference

### Key Shortcuts

| Key | Action |
|-----|--------|
| Tab | Toggle between Plan and Build modes |
| @ | Reference files or mention agents |
| / | Execute commands |

### Common Commands

```bash
/init          # Initialize project
/connect       # Configure API providers
/share         # Share conversation
/undo          # Undo last change
/redo          # Redo undone change
/help          # Show help
```

### File Locations

| Purpose | Project | Global |
|---------|---------|--------|
| Configuration | `opencode.json` | `~/.config/opencode/config.json` |
| Agents | `.opencode/agents/` | `~/.config/opencode/agents/` |
| Commands | `.opencode/commands/` | `~/.config/opencode/commands/` |
| Plugins | `.opencode/plugins/` | `~/.config/opencode/plugins/` |
| Tools | `.opencode/tools/` | `~/.config/opencode/tools/` |
| Auth | - | `~/.local/share/opencode/auth.json` |

### Troubleshooting

```bash
# Verify credentials
opencode auth list

# Check configuration
cat opencode.json | jq .

# Test provider connection
/connect
```

---

For more information: https://opencode.ai/docs/