gptcommit Files

Upgrade notes

• Building from source now requires Rust 1.88 or newer.
• file_ignore now uses Git-style patterns instead of substring matching. Use *.lock for a filename suffix, /path for a repository-root path, and ordered ! rules to include exceptions. Review custom rules when upgrading; for example, a substring rule lock should become *.lock to exclude lockfiles. These rules are separate from the repository's .gitignore.
• The default OpenAI model is now gpt-6-luna, with low reasoning effort and a 2,048-token limit shared by reasoning and output for each request. Explicit model overrides are preserved and do not receive these model-specific options. Choose another model with gptcommit config set openai.model <model> if needed.

Improvements

• Configuration reads display booleans, numbers, and arrays correctly. Setting or deleting a key edits only that key in the selected file, preserves comments, and avoids copying environment or unrelated configuration overrides into it.
• Ignore rules support globs, directories, anchored paths, and negation. Verbose output identifies excluded files and the rules that matched them.
• Empty or fully ignored changes leave the commit message untouched and make no API calls. Routine progress output is quieter; use --verbose for details.
• File summarization runs at most four requests concurrently, preserves stable output order, and reports failures instead of silently omitting summaries.
• Default prompts request concise, factual titles and bullets without invented motivation or test claims. Bun and uv lockfiles are excluded by default.
• OpenAI client and dependency updates improve compatibility and retry handling; errors are reported without panics, with tokenization fallback for unknown models.

Release maintenance

• Updated GitHub Actions are pinned to commit hashes.
• Homebrew and Winget updates wait until all release binaries are uploaded, removing the race with archive downloads.