Skip to content

CHANGELOG.md

Latest commit

 

History

History
132 lines (88 loc) · 4.45 KB

CHANGELOG.md

File metadata and controls

132 lines (88 loc) · 4.45 KB
Raw

[Unreleased]

[0.3.12] - 2026-08-09

Fixed

  • Models that stream their chain of thought inside content (as <think>...</think>, or with only the closing tag) no longer leak drafts and commentary into the commit message
  • A message the model wraps in a code fence is unwrapped instead of committed with the backticks
  • Raised the streaming cutoff so the real message is no longer truncated when inline reasoning precedes it

Changed

  • The detailed prompt now carries a worked example and forbids thinking out loud, drafting, and trailing commentary
  • Requests also send chat_template_kwargs: {enable_thinking: false} alongside disable_reasoning, which SGLang/vLLM back ends honour; a provider that rejects it still falls back through the existing can_disable_reasoning retry

[0.3.11] - 2026-08-09

Added

  • Commit message detail level, chosen with aicm -d / aicm --detail and stored as commit_detail in config.yml: concise keeps the current subject-only message, detailed adds an explanatory paragraph and a bullet list of notable changes
  • aicm --help shows the configured detail level

Changed

  • The Edit action now opens the generated message in your editor via git commit -e, honouring GIT_EDITOR/core.editor/VISUAL/EDITOR, instead of asking for a single line of input

Fixed

  • git commit no longer goes through the shell, so a message containing backticks, quotes or $ is committed verbatim instead of being expanded or truncated

[0.3.10] - 2026-08-09

Added

  • Presets can declare a built-in model list, offered directly during setup with no detection request
  • Built-in model list for AMD Radeon (China), whose gateway always answers /models with 403
  • Presets can record each model's context window; setup reports it and suggests a matching diff_len instead of always defaulting to 32768

Fixed

  • Crash (undefined method 'empty?' for nil) when pressing Enter at the API key prompt to keep the existing key

[0.3.9] - 2026-08-08

Added

  • The model list fetch timeout is now configurable: aicm setup asks for it as its first step and stores it as model_fetch_timeout in config.yml, applying to aicm -p and aicm -m as well
  • aicm --help shows the configured timeout

[0.3.8] - 2026-08-08

Changed

  • Increased the model list fetch timeout from 5s to 30s, so slower providers can finish responding during setup

[0.3.7] - 2026-08-08

Added

  • AMD Radeon (China) provider preset (https://developer.amd.com.cn/radeon/api/v1), an OpenAI-compatible endpoint
  • README section for AMD Radeon with the token factory registration link and available models
  • README section for Apple Local Models via apple-to-openai

Fixed

  • ConfigManager.update_provider silently dropped writes for providers missing from an existing config.yml; newly added presets are now appended with their default base_url

[0.3.6] - 2026-02-27

Changed

  • Optimized commit message synthesis logic
  • Renamed commit_ai_no_model for clarity

Fixed

  • Corrected the "no changes" message display

[0.3.5] - 2026-02-04

Added

  • Commit message format option: simple, conventional and gitmoji

Fixed

  • Release script corrections and Homebrew formula update for v0.3.4

[0.3.4] - 2026-01-30

Added

  • Anthropic Claude and NVIDIA NIM provider presets; provider list reordered

Changed

  • Extracted diff handling into a new DiffHelpers module
  • Improved CLI UX with better padding and formatting

[0.3.3] - 2026-01-28

Changed

  • Modernized configuration handling and quote style
  • Updated dependencies
  • Added release prep and publish guides

[0.3.2] - 2026-01-28

Fixed

  • Crash caused by the color helper method

[0.3.1] - 2026-01-28

Fixed

  • Gemspec executable path so the Homebrew tarball build works
  • Executable name in the Homebrew formula

[0.3.0] - 2026-01-28

Added

  • YAML configuration system (~/.config/commitgpt/) with multi-provider support
  • Homebrew distribution via brew tap
  • Streaming responses and improved error handling
  • Clean working tree check
  • Interactive prompt to stage all changes when the diff is empty

Fixed

  • Ctrl+C interrupt handling

[0.2.0] - 2026-01-09

Added

  • Custom API endpoint support and model listing
  • Alternative response field support for non-OpenAI providers

Changed

  • Improved API error handling

[0.1.2] - 2023-02-15

  • Documentation updates

[0.1.1] - 2023-02-15

  • Maintenance release

[0.1.0] - 2023-02-14

  • Initial release