Releases
v0.2
Compare
Sorry, something went wrong.
No results found
What's New in v0.2
Streaming Writer Output
Real-time essay generation : Watch the essay being written word-by-word instead of waiting 30-60 seconds
Auto-scrolling to follow new content as it streams in
"Writing..." indicator with spinner during generation
New UI Controls
Regenerate button : Test new writing prompts on existing sessions without restarting
Cancels any ongoing generation
Deletes cached draft and generates fresh
Home button : Quick navigation back from the draft view at any time
Writer Agent Improvements
Refined writing constraints for more natural, human prose
Better voice preservation from interview transcripts
Improved paragraph structure and flow
Technical Improvements
Streaming support added to LLMClient protocol (OpenAI and Anthropic adapters)
Task cancellation handling for clean regeneration workflow
Unified logging and hardened LLM flows
Fixed research result schema issues
Bug Fixes
Fixed API key validation caching to avoid double keychain prompts
Various stability improvements
You can’t perform that action at this time.