We've been shipping features faster than writing these release posts—classic developer move, right? 😅Here's what you might have missed across our latest releases.
TL;DR: Key Points Summary
Gemini CLI Provider: Direct integration with Google's Gemini CLI
Claude Code Provider: Much expected direct integration!
Cerebras API Provider: New AI provider option for faster inference
Task Timeline Visualization: Visual overview of your conversation flow
Context Window Warning: Visual indicator when approaching token limits
Multi-File Operations (incl. edits!): Better handling of complex code changes
Copy Prompt Button: A convenient way to copy your prompt to the clipboard
Mermaid Diagram Auto-Fix: AI-powered syntax error correction for diagrams
MCP Tool Management: Enhanced Model Context Protocol integrations
Browser Tool Fixes: Fixed potential memory leak in Browser Tool
Improved Qdrant support: Multiple improvements for better code indexing
That's a lot, right?!
Noteworthy Changes
Task Timeline Visualization
Available in: v4.39.0 | Contributor: @hassoncs
Enhanced in: v4.43.0
Finally—a visual way to navigate your chat history. The timeline displays your conversation flow and allows you to click to jump to specific moments. It’s perfect for understanding what happened during long coding sessions.
It’s enabled by default, but you can disable it in Display Settings. In v4.43.0, we added colorful gutters that connect timeline events to actual chat messages.
Workspace Root Environment Variable
Available in: v4.40.0 | Contributor: @hassoncs
Terminal sessions now automatically include $WORKSPACE_ROOT pointing to your current workspace. No more navigating back to root for every command:
cd $WORKSPACE_ROOT && npx jest
Especially useful in deeply nested directories or multi-workspace setups.
Claude Code Provider Support
Available in: v4.44.0 | Contributor: @kevinvandijk | via Roo, inspired by Cline
Much requested direct integration with Claude Code!
Gemini CLI Provider Support
Available in: v4.44.0 | Contributor: @kevinvandijk | via Roo, inspired by Cline
Direct integration with Google's Gemini CLI for seamless local-to-cloud AI workflows.
Cerebras API Provider
Available in: v4.39.0 | Contributor: @markijbema | via Cline
Added support for Cerebras' lightning-fast inference API. Because sometimes you need your AI responses yesterday.
Developer Experience Improvements
Multi-Language Localization
Available in: v4.43.0 | Contributor: @chrarnoldus
We had a massive wave of new users from the Philippines and Thailand! Of course, added support for them, as well as for the Ukrainian, Czech, Greek, and Swedish languages. Because good tools should speak your language!
Is that your native language? Help the community by reading through the files, and let us know if we did well. We don’t have native speakers in the core team!
Context Window Progress Bar Enhancement
Available in: v4.43.0 | Contributor: @hassoncs
The progress bar now turns red when you're approaching token limits. No more mysterious context window errors—you'll see them coming.
Mermaid Diagram Auto-Fix
Available in: v4.43.1 | Contributor: @markijbema
Broken Mermaid diagram? Hit the fix button and let AI correct the syntax errors. Because debugging diagram syntax is nobody's favorite hobby.
Enhanced Multi-File Operations
Available in: v4.39.0 | Contributors: @samhvw8, @daniel-lxs | via Roo Code
Experimental multi-file edits for complex refactoring
Better diff error handling and UI feedback
Fixed multi-file diff operations with proper start line handling
Improved diff editor race conditions in multi-monitor setups
Copy Prompt Button
Available in: v4.45.0 | Contributors: @Juice10 | via Community @vultrnerd
A convenient way to copy your prompt to the clipboard!
Major Roo Code Integrations
Roo Code v3.20.3 Integration
Available in: v4.39.0 | Contributor: @kevinvandijk
This massive merge brought dozens of improvements from the Roo codebase:
Security & Workspace Management:
Limit search_files to only look within workspace for improved security
Prevent auto-approving edits of configuration files
Respect auto-approve settings when searching/listing files outside workspace
Force tar-fs >=2.1.3 for security vulnerability fix
Performance & Memory:
Replace explicit caching with implicit caching to reduce latency for Gemini models
Lower concurrent file read limit default to 5 files
Add cache breakpoints for custom vertex models on Unbound
Move concurrent reads setting to context settings
Developer Experience:
Experimental multi-file edits support
LaTeX math equation rendering in chat window
Prompt history navigation via arrow up/down
Indexing status badge in chat view
Mermaid diagram support buttons
Model & Provider Improvements:
Add DeepSeek R1 support to Chutes provider
Add reasoning budget support to Bedrock models
Update XAI models and pricing
Add manual OpenAI-compatible format specification and parsing
Roo Code v3.21.5 Integration
Available in: v4.42.0 | Contributor: @kevinvandijk
Another substantial merge focusing on stability and usability:
Database & Search:
Fix Qdrant URL prefix handling for QdrantClient initialization
Resolve URL port handling bug for HTTPS URLs in Qdrant
Improve LM Studio model detection to show all downloaded models
UI & UX Improvements:
Resolve diff editor issues with markdown preview associations
Fix start line not working in multiple apply diff
Add LaTeX math equation rendering in chat window
Onboarding tweaks to emphasize modes
Rename 'Boomerang Tasks' to 'Task Orchestration' for clarity
Provider & Integration Fixes:
Add Claude Code provider for local CLI integration
Mark unused Ollama schema properties as optional
Fix context length for lmstudio and ollama
Resolve Claude Code provider JSON parsing and reasoning block display
Performance & Stability Fixes
MCP (Model Context Protocol) Improvements
Available in: v4.39.0, v4.42.0 | Contributors: @samhvw8, @NamesMT, @daniel-lxs, @Rexarrior | via Roo Code
Magic variables support with workspaceFolder injection
Toggle for excluding MCP server tools from prompts
Better execution UX and response handling
Enhanced marketplace display and timeout fixes
MCP tool eye icon state fixes
Browser Tool Fixes
Available in: v4.38.1, v4.42.0 | Contributors: @markijbema
Properly close browser tools when remote browsers are configured but local fallbacks are used
Fixed OpenRouter endpoint CSP errors when loading from custom URLs
Context and Navigation Improvements
Available in: v4.39.0 | Contributors: @daniel-lxs, @KJ7LNW, @SECKainersdorfer | via Roo Code
Prompt history navigation via arrow keys (finally!)
Support for escaping context mentions
Lower context condensing thresholds for better performance
Profile-specific context condensing settings
Model Provider Updates
Available across multiple releases | Contributors: @daniel-lxs, @edwin-truthsearch-io, @chrarnoldus | via Roo Code
Updated XAI models and pricing
Updated O3 model pricing
Correct Gemini 2.5 Flash pricing
OpenRouter BYOK cost calculation fixes
LM Studio model detection improvements
DeepSeek R1 support via Chutes provider
Reasoning budget support for Bedrock models
Polish & Bug Fixes
UI Improvements
Available in: v4.40.1 | Contributors: @hassoncs, @chrarnoldus
Fixed task timeline layout shift from hidden scrollbar
Fixed TelemetryService initialization error when diff apply fails (TelemetryService is a Roo leftover and disabled in Kilo)
Cache and Performance Fixes
Available in: v4.39.2, v4.44.1 | Contributor: @chrarnoldus
Fixed cache read stats not being displayed in the chat window. Because knowing your cache hit rate actually matters for performance debugging.
Fixed crashes with the error message "Bad substitution" and "Cannot read properties of undefined (reading 'includes')"
Terminal and File System Enhancements
Available in: v4.39.0, v4.42.0 | Contributors: @Ruakij, @village-way | via Roo Code
File write tool validation improvements
Fixed recursive directory scanning in @ mention "Add Folder" functionality
Symlink support in list_files tool
Terminal busy state reset fixes
Get the Latest Release
Download the latest version!
Want to stay updated? Follow us at kilocode.ai/social for the latest news and updates.
P.S. - We promise to write these release digests faster than we ship features. Eventually.