What's new?

Undo changes (revert entire project tree back to a checkpoint)
Completely reworked how chat context is built to align with the format expected by LLMs.
This resolves multiple issues, including:
1. Infinite looping
2. Invalid tool call formats
3. Unnecessary, never-ending improvements suggested by Sonnet
Replaced Diffs with the experimental "LLM Apply" — a cursor-style approach for applying code changes to existing files.
Most of the models struggle to produce correct diff patches.
Now LLM can suggest code changes in a natural language format and another cheaper and faster LLM will apply these changes to file.
Suggested code changes are currently applied by Groq.ai Cloud (Llama 70B) at incredibly high speeds!
This new method eliminates syntax errors, reduces token usage, and improves accuracy.
While also allowing more models to be used for coding because of much simpler output format.
Integrated Voyage.ai’s voyage-code-3, a state-of-the-art embeddings model specifically designed for code.
Better embeddings enhances code search, related file suggestions, and documentation retrieval.
o3-mini
UI theme color and code highlighting updates
All links (including links in terminal) open in built in browser
Cache improvements, reduction of token usage and minor bug fixes
Performance optimizations

Add individuals files to chat context
Mention the filename and click on suggested files to add them to context
Add all changed files tracked by Git with a single click to chat context
Need someone to review your code changes?
Add all changed files to chat with one click. Include diff of changes and ask for code review or help understanding why your changes don't work

Research
Multiple AI agents will perform research in parallel to find all relevant information about the task and project. Helps AI models understand the project better and work on existing codebases
Planner
Creates a plan before task execution and shows the progress
File context management
Helps reduce token usage by allowing you to remove no longer needed files from chat context
Automatic code linter runs after file changes
Automated code validation provides feedback to AI model when code validation fails for:
JavaScript, Python, Java, TypeScript, C, C++, C#, Ruby, Go, PHP, Rust, HTML, JSON, CSS

Integrated real Browser with Dev Tools - (resolves console errors automatically, takes screenshots)
Git integration
New UI
Minor bug fixes

Claude Sonnet 3.5
gpt-4o-mini
"diff" preview for code changes
reduced token usage ( multiple file operations in one tool call help reduce number of roundtrips)

Latest GPT-3.5 Model
Basic GPT-4 Vision Support
Task-based conversation
Dynamic Context (dynamic system message, summarization of conversation when exceeds certain limit) - reduces tokens usage significantly

File names are clickable, configure to open your IDE of choice by clicking on a file name
Use "arrow up" key to bring up last message from chat

Semantic code search. Vector embeddings created for source code files in the project with OpenAI API embeddings. All data and embeddings are stored locally on your computer.
Ask questions about your code, or let AI decide when it should use code search.
Code search makes it easier to work on existing projects

Smart Context - dynamic context that helps reduce token usage and extend context limitations. This feature will get improved over time.

Chat history. Save entire chat and continue from where you left off.
Also can be used as a way to build and save reporting dashboards in SQL mode.

All chat history is saved on your computer

Copy, Delete message functionality
Allows deleting all conversation down from a certain point. Also can delete attached files
SQL - Refresh button was added, now you can rerun SQL query
Code - Ability to read multiple files at once - saves tokens usage and time by reducing round trips
Code - Ability to make multiple updates at ounce - saves tokens usage
What's New link
Other minor UI/UX improvements