aicommit2

The codeReview feature performs an automated code review before generating the commit message.

CAUTION: This is an experimental feature. It significantly increases processing time, token consumption, and cost.

To enable it:

aicommit2 config set codeReview=true

You can also specify a custom prompt for the review using codeReviewPromptPath.

aicommit2 config set codeReview=true
aicommit2 config set codeReviewPromptPath="/path/to/user/prompt.txt"

aicommit2 resolves settings using a specific hierarchy. If a setting is defined in multiple places, the one higher in this list takes precedence:

1. Command-line arguments (e.g., --OPENAI.locale="jp")
2. Environment variables (e.g., OPENAI_API_KEY=...)
3. Configuration file (config.ini)
4. Model-Specific Settings (settings within a provider section like [OPENAI])
5. General Settings (global settings like temperature)
6. Default values

OpenRouter supports advanced features like structured output and reasoning-specific controls through specific payload objects:

• Structured Output: To ensure the model returns structured data, set OPENROUTER.responseFormat to a JSON object (e.g., {"type":"json_object"}).
• Routing Control: For more granular control over how OpenRouter routes your request, set the OPENROUTER.provider object directly in your configuration.
• Reasoning: To use reasoning-specific controls, set the OPENROUTER.reasoning object.

If you encounter issues where a model does not support your chosen format, run aicommit2 doctor to check catalog reachability and receive suggestions for compatible responseFormat or reasoning options.

The tool operates by running a diff command (e.g., git diff) to capture your latest code changes. This diff is then sent to your configured AI provider, which generates a structured commit message based on the changes.

Troubleshooting Large Diffs: If your diff is too large, the AI may fail to process it or return invalid messages. If you encounter errors related to message length or validity, try reducing the size of your commit unit (i.e., commit fewer files at once).

There are two ways to authenticate with the COPILOT_SDK provider:

1. Copilot CLI (Recommended)

This is the easiest method and requires no environment variables. Run the following commands in your terminal:

1. copilot
2. /login (to authenticate)
3. /model (to select your preferred model)

2. Token-based Authentication (Advanced/CI)

For headless or CI environments, use a Fine-Grained PAT (github_pat_...) with the Copilot Requests permission.

• Set the COPILOT_GITHUB_TOKEN environment variable.
• Note: Classic PATs (ghp_...) are not supported.
• aicommit2 isolates this auth: it uses COPILOT_GITHUB_TOKEN and ignores GH_TOKEN or GITHUB_TOKEN for COPILOT_SDK requests.

# For CI/Headless environments
export COPILOT_GITHUB_TOKEN=github_pat_...

To use GitHub Models with aicommit2, you can either use the recommended automatic login via the GitHub CLI or manually configure a Personal Access Token (PAT).

Option 1: Automatic login (Recommended)

This method authenticates using the GitHub CLI, stores the token in GITHUB_MODELS.key, and verifies your access.

aicommit2 github-login

Prerequisites: You must have the GitHub CLI installed on your system.

Option 2: Manual token setup

If you prefer using a PAT, ensure the token has the models: read permission. You can then configure it using the github-login command with the --token flag or by setting the config key directly.

aicommit2 github-login --token github_pat_xxxxxxxxxxxxxxxxxxxx
# or
aicommit2 config set GITHUB_MODELS.key="github_pat_xxxxxxxxxxxxxxxxxxxx"

Starting from Ollama v0.1.33 and aicommit2 v1.9.5, you can make simultaneous requests to multiple models using Ollama's experimental parallel loading feature.

1. Start the Ollama server with the OLLAMA_MAX_LOADED_MODELS environment variable set to the number of models you want to load:

   OLLAMA_MAX_LOADED_MODELS=3 ollama serve

2. Configure aicommit2 to use a comma-separated list of models:

   aicommit2 config set OLLAMA.model="mistral,dolphin-llama3"

3. Run aicommit2 as usual.

For CI/CD or security, you can provide API keys using environment variables. These take precedence over the configuration file. The standard variable names are:

• OPENAI_API_KEY
• ANTHROPIC_API_KEY
• GEMINI_API_KEY
• MISTRAL_API_KEY
• CODESTRAL_API_KEY
• COHERE_API_KEY
• GROQ_API_KEY
• PERPLEXITY_API_KEY
• DEEPSEEK_API_KEY

You can customize which environment variable a provider looks for by setting the envKey property in your configuration.

OPENAI_API_KEY="your-openai-key" ANTHROPIC_API_KEY="your-anthropic-key" aicommit2

You can use your locally installed Claude Code CLI as a provider for aicommit2. This allows you to use your Claude Pro/Max subscription for generating commit messages without needing an Anthropic API key.

Requirements

• Claude Code CLI installed: Install it globally via npm: npm install -g @anthropic-ai/claude-code.
• Authentication: You must run claude at least once in your terminal and complete the login process.
• Subscription: You must have an active Claude Pro or Max subscription, as each request consumes your subscription quota.

Configuration

This provider is opt-in. It remains inactive until you explicitly configure a model. You can configure it via the CLI or by editing your configuration file.

To activate it using the CLI, set the model (e.g., sonnet, opus, or haiku):

aicommit2 config set CLAUDE_CODE.model=sonnet