gcop-rs

Provider Configuration ​

gcop-rs supports multiple LLM providers. You can use built-in providers or add custom ones.

gcop-rs does not hardcode model allowlists. The examples below are common choices, not an exhaustive or validated list.

Streaming compatibility ​

Claude, OpenAI Chat Completions, OpenAI Responses, and Gemini use SSE when llm.stream_transport is enabled. Their decoders accept fields with or without a space after data:, LF/CRLF/CR line endings, multiline data, and UTF-8 characters split across HTTP chunks. This also applies to compatible models such as Kimi served through a gateway.

Each SSE event ends with a blank line. A stream must reach its protocol's completion signal: message_stop for Claude, [DONE] or finish_reason for Chat Completions, response.completed or [DONE] for Responses, and finishReason for Gemini. An EOF before that signal remains a truncation error.

Built-in Providers ​

Claude (Anthropic) ​

toml
[llm.providers.claude]
api_key = "sk-ant-your-key"
model = "claude-sonnet-4-5-20250929"
temperature = 0.3
max_tokens = 2000

Get API Key: https://console.anthropic.com/

Example Models:

  • claude-sonnet-4-5-20250929 (recommended)
  • claude-opus-4-5-20251101 (most powerful)
  • claude-3-5-sonnet-20241022 (older version)

OpenAI ​

toml
[llm.providers.openai]
api_style = "openai"  # optional for this built-in provider; inferred from the name
api_key = "sk-your-openai-key"
model = "gpt-4o-mini"
temperature = 0.3

Get API Key: https://platform.openai.com/

Use api_style = "openai-response" for OpenAI models that require or work better with the Responses API:

toml
[llm.providers.openai-response]
api_style = "openai-response"
api_key = "sk-your-openai-key"
endpoint = "https://api.openai.com"
model = "gpt-4o-mini"

In that mode, gcop-rs sends the system prompt as instructions, the user prompt as input, maps max_tokens to max_output_tokens, disables tool calls with tool_choice = "none", and parses output_text content from the response.

Example Models:

  • gpt-4o-mini (matches the built-in CI default)
  • gpt-4o
  • Any Chat Completions or Responses compatible model from OpenAI

Ollama (Local) ​

toml
[llm.providers.ollama]
endpoint = "http://localhost:11434"
model = "llama3.2"

Setup:

bash
# Install Ollama
curl https://ollama.ai/install.sh | sh

# Pull a model
ollama pull llama3.2

# Start server
ollama serve

Example Models: Any model available in Ollama (llama3.2, qwen2.5-coder, deepseek-coder-v2, etc.)

Gemini (Google) ​

toml
[llm.providers.gemini]
api_key = "AIza-your-gemini-key"
model = "gemini-3-flash-preview"
temperature = 0.3

Get API Key: https://ai.google.dev/

Example Models:

  • gemini-3-flash-preview (recommended default)
  • gemini-2.5-flash
  • gemini-2.5-pro

Custom Providers ​

You can add OpenAI-, Claude-, or Gemini-compatible APIs using the api_style parameter.

DeepSeek ​

toml
[llm.providers.deepseek]
api_style = "openai"
api_key = "sk-your-deepseek-key"
endpoint = "https://api.deepseek.com/v1/chat/completions"
model = "deepseek-chat"
temperature = 0.3

Get API Key: https://platform.deepseek.com/

Qwen (通义千问) ​

toml
[llm.providers.qwen]
api_style = "openai"
api_key = "sk-your-qwen-key"
endpoint = "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions"
model = "qwen-max"

Claude Proxy/Mirror ​

toml
[llm.providers.claude-code-hub]
api_style = "claude"
api_key = "your-key"
endpoint = "https://your-claude-code-hub.com/v1/messages"
model = "claude-sonnet-4-5-20250929"

Custom OpenAI Compatible Service ​

toml
[llm.providers.my-llm]
api_style = "openai"
api_key = "your-key"
endpoint = "https://api.example.com/v1/chat/completions"
model = "custom-model"

API Style Parameter ​

The api_style parameter determines which API implementation to use:

ValueDescriptionCompatible Services
"openai"OpenAI Chat Completions APIOpenAI, DeepSeek, Qwen, most custom services
"openai-response"OpenAI Responses APIOpenAI Responses API
"claude"Anthropic Messages APIClaude, Claude proxies/mirrors
"ollama"Ollama Generate APILocal Ollama only
"gemini"Google Gemini GenerateContent APIGemini and Gemini-compatible endpoints

If api_style is not specified, it defaults to the provider name (for backward compatibility with built-in providers).

Endpoint Rules ​

  • Claude, OpenAI, and Ollama providers accept either a base URL or a full request path in endpoint.
  • For OpenAI Responses API, use api_style = "openai-response"; a base endpoint = "https://api.openai.com" will resolve to /v1/responses.
  • Gemini expects a base URL in endpoint; gcop-rs derives /v1beta/models/{model}:generateContent from that base.

Thinking Tags ​

Some OpenAI-compatible models return visible reasoning in <thinking>...</thinking> or <think>...</think> blocks. gcop-rs preserves those blocks by default. To strip them from generated commit messages and review JSON before parsing, enable:

toml
[llm.providers.openai]
strip_thinking = true

Switching Providers ​

Using Command-Line ​

bash
# Use different provider for one command
gcop-rs --provider openai commit
gcop-rs --provider deepseek review changes

Changing Default ​

Edit your platform-specific config file (see Configuration Guide):

toml
[llm]
default_provider = "deepseek"  # Change this

API Key Management ​

Config File ​

Provider api_key is configured in config.toml:

toml
[llm.providers.claude]
api_key = "sk-ant-..."

CI Mode Environment Variables ​

In CI mode (CI=1), use environment variables instead of config file:

  • GCOP_CI_PROVIDER - Provider type: claude, openai, ollama, or gemini
  • GCOP_CI_API_KEY - API key
  • GCOP_CI_MODEL (optional, has defaults)
  • GCOP_CI_ENDPOINT (optional)

See Also ​