OpenCode Go CLI

Use OpenCode Go or OpenAI models with Claude Code.

Translates Claude Code's Anthropic requests into OpenAI API calls — in real time, chunk by chunk. Supports both OpenCode Go (API key) and OpenAI via OAuth (ChatGPT Plus/Pro).

Why OpenCode Go CLI

Claude Code uses the Anthropic API natively, but OpenCode Go offers coding models through an OpenAI-compatible API. This CLI bridges the two:

Spawns a local HTTP proxy that translates Anthropic requests to OpenAI format
Streams responses back in real time (no buffering)
Spawns Claude Code with the correct environment variables pointing to the local proxy
Persists your API key and last model between sessions

The goal is not to change how Claude Code works. The goal is to make it work with OpenCode Go — transparently.

Core Capabilities

Streaming translation — SSE chunk-by-chunk, in real time
Tool call support — tool_use and tool_result blocks are mapped correctly
Image support — image content blocks converted to data: URIs
Two providers — OpenCode Go (API key) or OpenAI via OAuth (ChatGPT Plus/Pro)
Interactive menu — Provider → Model → Permission Mode flow with settings menu
4 permission modes — default, acceptEdits, auto, bypassPermissions
WebSearch interception — web_search server tool requests executed locally via SearXNG Docker container
Model selection — interactive or --model flag, 10 models available
Persistent config — API key / OAuth tokens and last model stored in ~/.opencode-go-cli/
Sub-agent compatibility — all sub-agents use the same model via environment variables
Fail-fast validation — invalid models caught before proxy starts
Automatic token refresh — OAuth tokens renewed automatically

Interactive Menu

When launched without arguments, the CLI shows an interactive menu:

opencode-go
  → "What do you want to do?" → Start / Settings
    → Start:
        → Select provider (OpenCode Go / OpenAI)
        → Auth check (API key or OAuth)
        → Select model
        → Select permission mode
        → Start proxy + launch Claude Code
    → Settings:
        → Set API key / Login OpenAI / Logout OpenAI / Reset all

Permission Modes

Mode	Behavior
`default`	Asks permission for everything
`acceptEdits`	Auto-approves file edits, asks for commands
`auto`	Classifier reviews actions (experimental)
`bypassPermissions`	Skips all permission checks

Runtime Model

The CLI runs in two stages:

Stage 1: CLI (opencode-go)
  └─ Interactive menu (provider → model → permission mode)
  └─ Starts proxy on localhost:PORT
  └─ Starts SearXNG Docker container (background, for WebSearch)
  └─ Spawns Claude Code with ANTHROPIC_BASE_URL=http://localhost:PORT

Stage 2: Proxy (localhost:PORT)
  └─ Receives POST /v1/messages (Anthropic format)
  └─ Intercepts web_search server tool requests → SearXNG
  └─ Routes based on provider:
       ├─ OpenCode Go: Chat Completions API (/v1/chat/completions)
       └─ OpenAI:      Responses API (/backend-api/codex/responses)
  └─ Translates response back (streaming or non-streaming)
  └─ Returns to Claude Code

Request Flow

OpenCode Go (Chat Completions):

Claude Code → POST /v1/messages (Anthropic)
  → convertAnthropicRequestToOpenAI()
    → OpenCode Go API: POST /v1/chat/completions
      → streamOpenAIToAnthropic() or convertOpenAIResponseToAnthropic()
        → Claude Code

OpenAI/Codex (Responses API):

Claude Code → POST /v1/messages (Anthropic)
  → convertAnthropicRequestToResponses()
    → Codex backend: POST /backend-api/codex/responses
      → streamResponsesToAnthropic() or convertResponsesApiToAnthropic()
        → Claude Code

Environment Variables Injected

Variable	Value
`ANTHROPIC_BASE_URL`	`http://localhost:PORT` (local proxy)
`ANTHROPIC_AUTH_TOKEN`	Your OpenCode Go API key
`ANTHROPIC_MODEL`	Selected model
`CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC`	`1`
`CLAUDE_CODE_SUBAGENT_MODEL`	Same as selected model
`ANTHROPIC_DEFAULT_MESSAGES_MODEL`	Same as selected model

Architecture

src/
├── index.ts                 Thin entry point (shebang + main export)
├── cli.ts                  Interactive menus, permission modes, Claude Code spawn
├── constants.ts            MODELS, OPENAI_MODELS, PROVIDERS, OAuth constants
├── config.ts               getConfig, saveConfig, deleteConfig
├── path.ts                 resolveClaudePath (claude binary in PATH)
├── env.ts                  buildClaudeEnv, cleanupClaudeCodeVars
├── logger.ts               createLogger (DEBUG/INFO/WARN/ERROR) + silenceLogger() + file logging
├── auth/
│   ├── oauth.ts           PKCE, exchange, refresh, JWT decode
│   └── server.ts          Local HTTP server for OAuth callback
├── search/
│   └── searxng.ts         SearXNG Docker container management + search queries
└── proxy/
    ├── types.ts                          Config, Model interfaces
    ├── helpers.ts                        mapStopReason, generateMsgId, convertImageSource, makeSSE
    ├── websearch-interceptor.ts          WebSearch server tool interception via SearXNG
    ├── request-conversion.ts             Anthropic → Chat Completions (OpenCode Go)
    ├── response-conversion.ts            Chat Completions → Anthropic (OpenCode Go)
    ├── stream-conversion.ts              Chat Completions SSE → Anthropic SSE (OpenCode Go)
    ├── request-conversion-responses.ts   Anthropic → Responses API (OpenAI/Codex)
    ├── response-conversion-responses.ts  Responses API → Anthropic (OpenAI/Codex)
    ├── stream-conversion-responses.ts    Responses API SSE → Anthropic SSE (OpenAI/Codex)
    └── server.ts                         Bun.serve + dual routing + WebSearch interception

tests/
├── helpers.test.ts
├── request-conversion.test.ts
├── response-conversion.test.ts
├── env.test.ts
└── logger.test.ts

Models Available

OpenCode Go (API key — `--provider opencode`)

ID	Name	Description
`minimax-m2.7`	MiniMax M2.7	High performance coding model
`minimax-m2.5`	MiniMax M2.5	Balanced speed and quality
`kimi-k2.5`	Kimi K2.5	Strong reasoning for complex tasks
`glm-5`	GLM-5	Latest generation from Zhipu AI

OpenAI (OAuth — `--provider openai`)

ID	Name	Description
`gpt-5.2`	GPT-5.2	Latest GPT-5 model
`gpt-5.3`	GPT-5.3	High performance GPT-5
`gpt-5.4`	GPT-5.4	Balanced GPT-5
`gpt-5.1-codex`	GPT-5.1 Codex	Code-optimized GPT-5.1
`gpt-5.2-codex`	GPT-5.2 Codex	Code-optimized GPT-5.2
`gpt-5.3-codex`	GPT-5.3 Codex	Code-optimized GPT-5.3

CLI

# Interactive mode (full menu: provider → model → permission mode)
opencode-go

# Direct launch with flags
opencode-go --model minimax-m2.7
opencode-go --provider openai --model gpt-5.2-codex
opencode-go --model minimax-m2.7 --permission-mode acceptEdits
opencode-go --provider openai --model gpt-5.2-codex --permission-mode auto
opencode-go --dangerously-skip-permissions --model minimax-m2.7

# Setup / Auth
opencode-go --setup           Configure OpenCode Go API key
opencode-go --oauth-login     Authenticate with OpenAI (ChatGPT Plus/Pro)
opencode-go --oauth-logout    Remove OpenAI tokens
opencode-go --reset           Delete all config

# List models
opencode-go --list                        Show default provider models
opencode-go --list --provider openai      Show OpenAI models
opencode-go --list --provider opencode    Show OpenCode Go models

# Proxy-only mode (for testing)
opencode-go --proxy --port 8080

# Interactive mode auto-selects the next free local proxy port if the preferred port is busy
opencode-go --model minimax-m2.7

# Help
opencode-go --help

Setup

Requirements:

Bun runtime (required to run opencode-go)
Claude Code installed and in PATH
OpenCode Go: API key from your OpenCode Go subscription
OpenAI: ChatGPT Plus/Pro subscription (for OAuth)

Install globally (recommended):

bun install -g @opencode-go/cli

O pacote é publicado no npm, mas o runtime oficial do CLI é Bun. O comando recomendado de instalação global é bun install -g @opencode-go/cli.

For local development only:

bun run build
bun link

Quick Start — OpenCode Go:

opencode-go --setup
opencode-go

Quick Start — OpenAI:

opencode-go --oauth-login
opencode-go --provider openai --model gpt-5.2-codex

Configuration

Config is stored at ~/.opencode-go-cli/config.json:

proxyPort is the preferred local proxy port. In normal interactive mode, if that port is already in use, the CLI automatically tries the next free port and injects the final URL into Claude Code.

{
  "provider": "opencode",
  "apiKey": "sk-opencode-...",
  "openaiTokens": {
    "access": "...",
    "refresh": "...",
    "expiresAt": 1234567890
  },
  "lastModel": "minimax-m2.7",
  "proxyPort": 8080
}

Development

opencode-go                   # After global install with Bun
bun run src/index.ts          # Without global install
bun run build                 # Build to dist/
bun run typecheck             # TypeScript check
bun test                      # Test suite (49 tests)

Proxy-only testing (with API key in config):

Interactive mode auto-falls back to the next free port. --proxy remains explicit/predictable: use --port to force a specific port, or omit it to use the configured/default preferred port.

# Without DEBUG (quiet)
opencode-go --proxy --port 8082

# With DEBUG (verbose)
DEBUG=1 opencode-go --proxy --port 8082

Test an invalid model (should fail fast):

opencode-go --model invalid-model
# → Unknown model: invalid-model
# → Run 'opencode-go --list' to see available models.

Documentation

Document	Purpose
CLAUDE.md	Claude Code guidance for this repository
.specs/project/ROADMAP.md	Milestones and feature status
.specs/codebase/ARCHITECTURE.md	Code architecture and data flows
.specs/codebase/STRUCTURE.md	File structure and module organization
.specs/codebase/INTEGRATIONS.md	External integrations and APIs

WebSearch (SearXNG)

The proxy intercepts web_search server tool requests from Claude Code and executes them locally via a SearXNG Docker container.

Container: opencode-searxng (port 8888)
Auto-started in background when proxy launches
Settings generated at ~/.opencode-go-cli/searxng/settings.yml
Requires Docker to be available; gracefully degrades if not

Current State

49 tests passing (Bun test runner)
Build clean (bun build → dist/index.js)
All 7 milestones complete:
- ✅ Modular refactoring (12 modules, ~40-220 lines each)
- ✅ Test suite (conversion, helpers, env, logger)
- ✅ Debug logging gated (DEBUG=1 for verbose output)
- ✅ UX improvements (spinner, fail-fast, clear errors)
- ✅ Codex OAuth provider (OpenAI via OAuth, GPT-5.x family)
- ✅ WebSearch interception via SearXNG
- ✅ Interactive CLI and permission modes

License

MIT

Name		Name	Last commit message	Last commit date
Latest commit History 10 Commits
.specs		.specs
src		src
tests		tests
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CLAUDE.md		CLAUDE.md
README.md		README.md
bun.lock		bun.lock
bunfig.toml		bunfig.toml
package.json		package.json
tsconfig.json		tsconfig.json

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

OpenCode Go CLI

Why OpenCode Go CLI

Core Capabilities

Interactive Menu

Permission Modes

Runtime Model

Request Flow

Environment Variables Injected

Architecture

Models Available

OpenCode Go (API key — `--provider opencode`)

OpenAI (OAuth — `--provider openai`)

CLI

Setup

Configuration

Development

Documentation

WebSearch (SearXNG)

Current State

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

OpenCode Go CLI

Why OpenCode Go CLI

Core Capabilities

Interactive Menu

Permission Modes

Runtime Model

Request Flow

Environment Variables Injected

Architecture

Models Available

OpenCode Go (API key — --provider opencode)

OpenAI (OAuth — --provider openai)

CLI

Setup

Configuration

Development

Documentation

WebSearch (SearXNG)

Current State

License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

OpenCode Go (API key — `--provider opencode`)

OpenAI (OAuth — `--provider openai`)

Packages