- handle null chunk.choices idk why
- fix tool description truncation for openai models #45
- support deepseek api reasoning model's reasoning content
- remove Anthropic from /models since it doesn't work yet
- fix max_tokens calculation
- add dynamic reasoning effort up to the defined max in /config or /model. You can say things like "megathink" and it'll be set to high. This follows claude-code and you can check out
src/utils/thinking.ts
for the details.
- fix lsTool to not show the safety warning to the user
- fix tool description truncation for openai models
- buffer the rawdog stream
- Rawdog the completion call because everyone sucks
- Add thinking and reasoning blocks to the assistant message
- Remove the call to updateTerminalTitle in REPL.tsx
- Get API key from env if available
- Add proxy support
- Fix #20
- Fix max token input field and reasoning effort in /config
- Fixed ! bash mode
- Fixed #3 An assistant message with 'tool_calls' must be followed by a tool result error
- Removed AutoUpdater component
- Added support for LLM servers that don't include
usage
or token counts in their responses (e.g. LM Studio)- Modified query handling to work with servers that don't return token usage metrics
- Improved compatibility with third-party LLM servers
- Updated issue templates for better bug reporting
- Updated README.md with improved documentation
- Implemented
/bug
command for submitting feedback directly from the CLI- Created Bug component to handle feedback submission workflow
- Integrated with GitHub issue reporting
- Updated issue templates for more structured bug reporting
- Fixed model selection and configuration bugs
- Resolved issues with model detection and selection
- Fixed configuration persistence issues
- Added support for OpenAI's
reasoning_effort
parameter- Updated models.ts to include
supports_reasoning_effort
flag for compatible models - Added configuration screen for max tokens and reasoning effort
- Modified the claude.ts service to respect these new parameters
- Updated models.ts to include
- Updated model config handling in API calls
- Fixed various bugs related to model configuration and token usage
- Updated README documentation with clearer usage instructions
- Added single release workflow that handles version bump, build, and publish
- Streamlined the release process with automated versioning
- Improved CI/CD pipeline
- Updated README with new single release workflow information
- Fixed version-bump workflow to properly trigger build and publish processes
- Fixed build process and dependency issues
- Improved build script reliability
- Fixed npm-publish workflow to install Bun for proper builds
- Improved deployment pipeline configuration
- Fixed npm-publish.yml workflow file for more reliable deployments
- Initial tagged release
- Basic CLI functionality established