Changelog
Stay up-to-date with the latest improvements and fixes in PURECODE AI.
Last updated
Stay up-to-date with the latest improvements and fixes in PURECODE AI.
Last updated
June 04, 2025
Dedicated Diff view to allow users to easily compare changes side-by-side in GUI section. Also users can seamlessly switch between the "Generated" code view and the new "Diff" view at any time after the initial generation is completed.
Replaced the 'Send' and 'Codebase' buttons with a single 'Send' button. Introduced a 'Use Codebase' checkbox (enabled by default), located next to the advanced indexing option below the chat area. When enabled, the context will be passed with the user query; if unchecked, the query will be sent without the context.
New scroll-to-bottom option added to help users quickly reach the bottom of the chat area. The button appears above the chat area in the center.
Floating edit option for recent input sections, enabling users to re-initiate requests directly from their previous input.
Custom client messages for better handling of status codes and responses.
Code snippet height adjustments while streaming, ensuring better readability and a smoother user experience.
UI enhancements for the Planning Section design, providing a more intuitive layout.
Updated the tool call UI, including the Read Codebase tool call UI.
We've updated the model dropdown to clearly show which models support chat and which don’t support agent tasks.
Fixed token and config refetching issues when using multiple windows of the extension instance.
Resolved UI wipeout issue with @Mentions in the chat area when users write a prompt and mentions provider in the middle, along with an extra space issue.
Fixed the issue where users weren’t receiving the full response inside the snippet when requesting documentation. Now, the entire response from the model is properly included inside the snippet, ensuring nothing is lost.
May 21, 2025
New tool calls in agent mode for reading problems and terminal issues, expanding the capabilities of agent mode.
New onboarding flow to guide users in discovering the full potential of PureCode AI extension features.
Custom model prompt tuning to handle edge cases for each model.
Back-and-forth chat with users. It enables interactive, humanized conversations for iterative coding support.
Checkpoint improvements for agent mode to enhance stability.
Stability and UI/UX improvements for a smoother user experience.
Complete refactoring of prompt architecture.
New prompt for all models.
Updated tool call with all functionality.
Preventing model from hallucinating and generating irrelevant content.
User message edge cases being handled (greeting messages, faulty messages, etc.).
Planning steps fixed for Gemini model.
May 10, 2025
Support for Checkpoint and Restore File Changes: This feature enables users to restore project files to specific snapshots, offering an easy way to roll back changes.
Agents for Codebase Context & Running Terminal Commands: Now you can manage the codebase context more efficiently and run terminal commands directly for streamlined workflows.
read_codebase Toolcall Implemented Using Provider \ A new toolcall method read_codebase has been introduced, implemented through a provider, offering enhanced access to the codebase for better management and context retrieval.
Trimming Content of Terminal Provider to 150 Lines: To keep things clean and efficient, the terminal provider will now limit its output to 150 lines. This helps prevent too much clutter and ensures better performance without overwhelming you with excessive data.
Preventing the Calculation of Advanced Indexing on Queries for Agent Mode: When working in agent mode, we've stopped the system from calculating advanced indexing during queries. This change improves speed and ensures the process is more efficient, without unnecessary tasks slowing things down.
Preventing Context Passing in Agent Mode: Context won't be automatically passed along in agent mode anymore. This makes sure each operation is more focused, avoiding any interference from previous contexts and keeping things precise.
Files Visible in Provider Once Created or Added to Codebase: Now, as soon as you create or add files to your codebase, they’ll be visible in the provider. No need to wait or refresh; you’ll have immediate access to your files when you need them.
More Dynamic Actions and Management for Generated Codes: The Accept All and Reject All options now appear above the Continue Generating button, even when responses are interrupted. Users can accept/reject all at once and continue the generation process, allowing them to manage further responses as they come. This applies to all assistant responses with code blocks for easier code management.
UI/UX Stability and Improvements: Removed gradient loaders and replaced them with a more straightforward "Generating" status view, including an option to stop the process. This update improves visual clarity and access during task generation.
Image Input Copy/Paste Issue Fixed: Resolved issues where image inputs were being pasted twice in the chat area and where no response was being received after pasting and hitting send.
May 01, 2025
Switched default model to Gemini 2.5 pro preview
UI/Ux Improvements
April 26, 2025
New Interactive Feature: Users can now smoothly navigate between different chat versions by hovering over horizontal dots in the chat area, making it easier to switch and view previous states of the conversation.
We've integrated support for three advanced AI models—GPT-4.1, Claude 3.7, and Gemini 2.5 Pro into our platform. These additions provide users with access to cutting-edge AI capabilities, offering enhanced performance, improved accuracy, and greater versatility across various tasks.
April 17, 2025
Removed the auth reducer from persistent storage.
Added an initial check in AuthWrapper from secretstorage to verify token availability. If the token is not present, it will attempt to fetch it from localStorage. If not available, the token will be removed from the Redux state.
Updated the design and interaction of the "Continue Generation" button.
Fixed an issue where the cursor pointer would jump while editing the planning step and prompt.
Resolved the issue with the missing advanced indexing loader during generation.
Fixed the reminder logic to notify users about advanced indexing based on their preferences stored in the database.
Resolved an issue where normal indexing was not being computed unless advanced indexing was enabled.
Users are now prevented from enabling advanced indexing while normal indexing is in progress. A toast notification is shown in such cases.
All indexing operations are now stopped on user sign-out and will automatically restart after sign-in.
Fixed inconsistencies in indexing status updates, particularly during Git operations and other edge cases such as RepoMapInd.
