Configuration Guide

Configuration Files

gcop-rs reads TOML configuration from two levels:

User-Level Config

This is your personal config (usually contains API keys):

Platform	Location
Linux	~/.config/gcop/config.toml
macOS	~/Library/Application Support/gcop/config.toml
Windows	%APPDATA%\gcop\config\config.toml

Project-Level Config (Optional)

Team-shared config in your repository:

Scope	Location
Project	<repo>/.gcop/config.toml

gcop-rs resolves the repository root by walking upward to the nearest .git boundary, then reads only <repo>/.gcop/config.toml at that root.

Effective Priority (High → Low)

1. CI overrides (CI=1 + GCOP_CI_*)
2. Environment overrides (GCOP__*)
3. Project-level config (.gcop/config.toml)
4. User-level config (platform-specific path above)
5. Built-in defaults

All config files are optional. Missing values fall back to lower-priority sources/defaults.

Quick Setup

Recommended: Use init commands

gcop-rs init
gcop-rs init --project   # optional: create .gcop/config.toml for team-shared settings

gcop-rs init creates your user-level config at the correct platform-specific location. gcop-rs init --project creates .gcop/config.toml at the current Git repository root.

Manual setup:

Linux:

mkdir -p ~/.config/gcop
cp examples/config.toml.example ~/.config/gcop/config.toml

macOS:

mkdir -p ~/Library/Application\ Support/gcop
cp examples/config.toml.example ~/Library/Application\ Support/gcop/config.toml

Windows (PowerShell):

New-Item -ItemType Directory -Force -Path "$env:APPDATA\gcop\config"
Copy-Item examples\config.toml.example "$env:APPDATA\gcop\config\config.toml"

Then edit the config file to add your API key.

Basic Configuration

Minimal configuration for Claude API:

[llm]
default_provider = "claude"

[llm.providers.claude]
api_key = "sk-ant-your-key-here"
model = "claude-sonnet-4-5-20250929"

Complete Configuration Example

# LLM Configuration
[llm]
default_provider = "claude"
# fallback_providers = ["openai", "gemini", "ollama"]  # Auto-fallback when main provider fails
max_diff_size = 102400  # Max diff bytes before truncation (commit/review/hook non-split flows)
stream_transport = true  # Use SSE for supported providers, even when output is buffered

# Claude Provider
[llm.providers.claude]
api_key = "sk-ant-your-key"
endpoint = "https://api.anthropic.com"
model = "claude-sonnet-4-5-20250929"
temperature = 0.3
max_tokens = 2000

# OpenAI Provider
[llm.providers.openai]
api_style = "openai"  # optional for this built-in provider; inferred from the name
api_key = "sk-your-openai-key"
endpoint = "https://api.openai.com"
model = "gpt-4o-mini"
temperature = 0.3
# strip_thinking = true      # Optional: remove <thinking>...</thinking> / <think>...</think> blocks

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

# Ollama Provider (local)
[llm.providers.ollama]
endpoint = "http://localhost:11434"
model = "llama3.2"

# Gemini Provider
[llm.providers.gemini]
api_key = "AIza-your-gemini-key"
model = "gemini-3-flash-preview"

# Commit Behavior
[commit]
show_diff_preview = true
allow_edit = true
split = false  # true = enable atomic split commit mode by default
max_retries = 10

# Optional commit convention guidance (prompt-level)
[commit.convention]
style = "conventional"  # conventional | gitmoji | custom
types = ["feat", "fix", "docs", "refactor", "test", "chore"]
template = "{type}({scope}): {subject}"  # useful with style = "custom"
extra_prompt = "Commit subject should be in English"

# Review Settings
[review]
min_severity = "info"  # critical | warning | info (applies to text output)

# UI Settings
[ui]
colored = true
streaming = true  # Render SSE deltas live in interactive commit flows
language = "en"  # Optional: force UI language (e.g., "en", "zh-CN")

# `stream_transport` controls HTTP SSE; `ui.streaming` only controls live rendering.
# Ollama does not support SSE and uses buffered responses.

# Network Settings
[network]
request_timeout = 120    # HTTP request timeout in seconds
connect_timeout = 10     # HTTP connection timeout in seconds
max_retries = 3          # Max retry attempts for failed API requests
retry_delay_ms = 1000    # Initial retry delay (exponential backoff)
max_retry_delay_ms = 60000  # Max retry delay; also limits Retry-After header

# File Settings
[file]
max_size = 10485760      # Max file size for `review file <PATH>` (10MB)
lockfile_patterns = ["**/*.lock"]  # Extra lockfile patterns summarized in LLM prompts

# Workspace Settings (monorepo scope inference)
[workspace]
enabled = true
members = ["packages/*", "apps/*"]  # Optional: override auto-detection
scope_mappings = { "packages/core" = "core", "packages/ui" = "ui" }

Configuration Options

LLM Settings

Option	Type	Default	Description
default_provider	String	"claude"	Default LLM provider to use
fallback_providers	Array	[]	Fallback provider list; automatically tries next when main provider fails
max_diff_size	Integer	102400	Maximum diff size (bytes) sent to LLM in commit/review/hook non-split flows; larger inputs are truncated
stream_transport	Boolean	true	Use HTTP SSE with supported providers, including JSON, hooks, split commits, and non-interactive flows that buffer the final response; set false only for endpoints that reject SSE

Provider Settings

Each provider under [llm.providers.<name>] supports:

Option	Type	Required	Description
api_style	String	No	API style: "claude", "openai", "openai-response", "ollama", or "gemini" (defaults to provider name if not set)
api_key	String	Yes*	API key used when a provider is instantiated or validated (*not required for Ollama)
endpoint	String	No	Custom endpoint/base URL. Claude/OpenAI/Ollama accept either a base URL or a full request path; Gemini expects a base URL because gcop-rs derives the final request path from model
model	String	Yes	Model name
temperature	Float	No	Temperature (0.0-2.0). Claude/OpenAI/Gemini-style defaults to 0.3; Ollama uses provider default when omitted
max_tokens	Integer	No	Max response tokens. Claude-style defaults to 2000; OpenAI-style sends only if set; Ollama currently ignores this field
strip_thinking	Boolean	No	Remove <thinking>...</thinking> and <think>...</think> blocks from generated commit/review text. Default is false
extra	Object	No	Additional provider-specific keys. Unknown keys are preserved; max_tokens/temperature are also read from here as a compatibility fallback

gcop-rs does not hardcode a model allowlist. Any model compatible with the selected API shape can be configured.

Commit Settings

Option	Type	Default	Description
show_diff_preview	Boolean	true	Show diff stats before generating
allow_edit	Boolean	true	Allow editing generated message
split	Boolean	false	Enable atomic split commit mode by default (same effect as always passing commit --split)
max_retries	Integer	10	Max generation attempts (including the first generation)
custom_prompt	String	No	Custom prompt instructions for commit generation (normal mode: replaces base commit system prompt; split mode: appended as additional grouping instructions)
convention	Table	No	Optional prompt-level convention guidance; see [commit.convention] below

Commit Convention Settings ([commit.convention])

These settings are prompt-level guidance for commit generation. They influence model output but are not hard validation rules.

Option	Type	Default	Description
style	String	"conventional"	Convention style: "conventional", "gitmoji", or "custom"
types	Array	No	Allowed commit types (mainly for conventional / custom)
template	String	No	Custom template hint (for example {type}({scope}): {subject})
extra_prompt	String	No	Additional plain-text instruction appended to convention guidance

Review Settings

Option	Type	Default	Description
min_severity	String	"info"	Minimum severity to display in text output: "critical", "warning", or "info"
custom_prompt	String	No	Custom system prompt / instructions for code review

UI Settings

Option	Type	Default	Description
colored	Boolean	true	Enable colored output
streaming	Boolean	true	Render received deltas live in interactive commit flows; does not control HTTP transport
language	String	null (auto)	Force UI language (e.g., "en", "zh-CN"); if unset, gcop-rs auto-detects

Legacy Keys: Older config files may still contain keys such as commit.confirm_before_commit, review.show_full_diff, or ui.verbose. These keys are currently ignored. Note on Streaming: llm.stream_transport controls HTTP SSE for OpenAI-, Claude-, and Gemini-style APIs. ui.streaming only controls whether interactive commit flows render those deltas as a typewriter effect. Ollama uses buffered responses regardless of either setting.

Network Settings

Option	Type	Default	Description
request_timeout	Integer	120	HTTP request timeout in seconds
connect_timeout	Integer	10	HTTP connection timeout in seconds
max_retries	Integer	3	Max retry attempts for failed API requests
retry_delay_ms	Integer	1000	Initial retry delay in milliseconds (exponential backoff)
max_retry_delay_ms	Integer	60000	Max retry delay in ms; also limits Retry-After header

File Settings

Option	Type	Default	Description
max_size	Integer	10485760	Max file size in bytes when using review file <PATH> (default: 10MB)
lockfile_patterns	Array	[]	Extra glob patterns for dependency lockfiles whose full diffs are never sent to the LLM

Common dependency lockfiles are built in and always sent as summary-only entries in commit/review/hook prompts, including Cargo.lock, package-lock.json, npm-shrinkwrap.json, yarn.lock, pnpm-lock.yaml, poetry.lock, Pipfile.lock, uv.lock, composer.lock, Gemfile.lock, go.sum, go.work.sum, bun.lockb, bun.lock, deno.lock, flake.lock, conan.lock, pubspec.lock, mix.lock, stack.yaml.lock, and Podfile.lock.

When a lockfile changes, gcop-rs still includes the file name and change counts such as Cargo.lock (+42 -7) [lockfile]; only the full patch body is omitted. This also applies when the lockfile is the only changed file.

Workspace Settings

Workspace settings control monorepo detection and commit scope inference.

Option	Type	Default	Description
enabled	Boolean	true	Enable workspace detection and scope inference
members	Array	No	Optional member patterns to use directly (skips auto-detection when set)
scope_mappings	Object	{}	Optional path-to-scope remap (for example "packages/core" = "core")

Auto-detection currently recognizes Cargo workspace, pnpm workspace, npm/yarn workspaces, Lerna, Nx, and Turborepo structures.

API Key Configuration

Sources

• User-level config file (platform-specific location, see above)
• Project-level config file (.gcop/config.toml, optional, for non-secret team settings)
• CI mode environment variables (GCOP_CI_*, only when CI=1)

When CI=1, CI-mode provider settings are applied after file/env loading, and become the effective default provider (ci).

Methods

Method 1: Config File (Recommended)

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

Method 2: CI Mode Environment Variables

export CI=1
export GCOP_CI_PROVIDER=claude
export GCOP_CI_API_KEY="sk-ant-your-key"

Security

Linux/macOS:

• Set file permissions: chmod 600 <config-file-path>

All platforms:

• Never commit your user-level config file (it may contain API keys)
• .gcop/config.toml is intended for team-shared non-secret settings and can be committed
• Do not put api_key in project-level config; use user-level config or environment variables instead

CI Mode

For CI/CD environments, gcop-rs provides a simplified configuration via environment variables. When CI=1 is set, you can configure the provider using GCOP_CI_* variables instead of a config file.

Required Variables

Variable	Description	Example
CI	Enable CI mode	1
GCOP_CI_PROVIDER	Provider type	claude, openai, openai-response, ollama, or gemini
GCOP_CI_API_KEY	API key	sk-ant-...

Optional Variables

Variable	Description	Default
GCOP_CI_MODEL	Model name	claude-sonnet-4-5-20250929 (claude) gpt-4o-mini (openai/openai-response) llama3.2 (ollama) gemini-3-flash-preview (gemini)
GCOP_CI_ENDPOINT	Custom API endpoint	Provider default

Example

#!/bin/bash
# CI workflow example

export CI=1
export GCOP_CI_PROVIDER=claude
export GCOP_CI_API_KEY="$SECRET_API_KEY"  # from CI secrets
export GCOP_CI_MODEL="claude-sonnet-4-5-20250929"

# Generate commit message
gcop-rs commit --yes

Benefits of CI Mode:

• No config file needed - all configuration via environment variables
• Provider name is automatically set to "ci"
• Simplifies GitHub Actions / GitLab CI integration
• Secrets can be injected via CI/CD secret management

Environment Overrides (GCOP__*)

In addition to CI-mode provider env vars, gcop-rs supports overriding configuration values via environment variables with the GCOP__ prefix.

• Priority: GCOP__* overrides config file and defaults.
• Mapping: Nested keys are separated by double underscores (__).
• Note: If CI=1, CI-mode provider settings are applied after this stage and become the effective default provider.

Examples:

# Disable colors and live streaming output
export GCOP__UI__COLORED=false
export GCOP__UI__STREAMING=false

# Disable HTTP SSE for endpoints that reject streaming requests
export GCOP__LLM__STREAM_TRANSPORT=false

# Switch default provider
export GCOP__LLM__DEFAULT_PROVIDER=openai

# Force UI language
export GCOP__UI__LANGUAGE=zh-CN

Locale Selection Priority

gcop-rs resolves UI language in this order:

1. GCOP__UI__LANGUAGE environment variable
2. [ui].language in config file
3. System locale
4. Fallback to English (en)

Override with Command-Line

# Override provider
gcop-rs --provider openai commit

# Enable verbose mode
gcop-rs -v commit

Command-line options override configuration file.

See Also

• Provider Setup - Configure LLM providers
• Provider Health Checks - Validation behavior and health endpoints
• Custom Prompts - Customize AI prompts
• Troubleshooting - Common configuration issues