March Changelog: New Features; MCP in Refact.ai [AMA Recording]; Community Use Cases with AI Agent
In the March edition of our Changelog:
- Latest updates in Refact.ai
- 93,3% score on Aider’s polyglot bench
- AMA recording: MCP in Refact.ai Agent
- Must-see new content: Vibe coding tips & AI Agent real use cases
Read on for all the details!
What’s new in Refact.ai in March?
💭 ‘Think’ button: A new dedicated button to activate models’ reasoning capabilities, improving problem-solving for complex tasks.
⏳ Daily usage tracking: View remaining daily requests to AI Agent in the Settings dropdown. By default, Free plan offers 8/day, Pro plan — 40/day.
📦 Smart context compression for longer chats: Conversations now support 2x more context, with automatic compression when reaching limits (32K Free / 64K Pro). Users get a notification when compression occurs and can reopen chat in a compressed format to restore context and limits.
🧩 Faster access to AI Agent integrations: A new button streamlines integration with GitHub, Docker, MCP servers, and more.
👀 MCP Server logs: Real-time diagnostic panel for monitoring server processes, tool execution, and connection status.
📚 Knowledge management (Experimental, VS Code): New interface for accessing and navigating stored knowledge. Enable in Extension settings → Xperimental to access ‘Manage Knowledge’ in the Settings bar. While this UI is only available in VS Code, knowledge still works in JetBrains.
New features are available in the latest plugin versions: 6.3.3 for VSCode, 6.3.0 for JetBrains.
Enterprise/OSS Updates
🆕 Model updates: Added Claude 3.7 Sonnet & o3-mini models (for OSS) and QwQ-32B, QwQ-32B-AWQ (4-bit quantized) models with tool use and reasoning (for Enterprise).
📤 Improved thinking models: Better handling of local reasoning process and responses.
Refact.ai Agent scores highest on Aider’s polyglot bench: 93.3% with Thinking Mode, 92.9% without
Refact.ai Agent, powered by Claude 3.7 Sonnet, has achieved top performance on Aider’s polyglot benchmark — through fully autonomous development.
Our AI Agent follows an iterative approach that requires no manual user input. It plans → executes → tests → self-corrects to deliver a fully working, accurate solution inside your IDE.
True vibe coding: you delegate, AI executes, and the result is ready!
Explore the blog post for our autonomy-first approach in detail, AI workflow comparisons, Thinking mode impact, and additional technical insights.
🔥 Learn how to use MCP in AI programming: Watch our recent AMA session
Curious how Model Context Protocol (MCP) enhances AI agents for coding? Catch our AMA session recording to explore:
- MCP explained: What it is and how it works
- Live demo: Setting up MCP in Refact.ai Agent for autonomous integration with hundreds of dev tools.
- Q&A: Real-world use cases breakdown.
Speakers: Oleg Klimov, Founder & CEO; Nick Frolov, Head of Product.
Thanks to everyone who joined us live! For future AMA announcements, follow us on X and LinkedIn.
Tips and guides to empower your AI programming
Ready to master vibe coding? Our new guide brings you 10 practical tips for effective AI-driven development while keeping high engineering standards. Complete with real cases, key considerations, and expert insights.
March community showcase: our Ambassadors shared remarkable stories of how they transform weeks of coding into minutes with Refact.ai Agent!
- How Refact.ai built 99.9% of an IoT cloud app [read]
- How AI-coded prototypes help UX teams deliver real value [read]
- 80 hours of building from scratch — instead done in 30 minutes [read]
- 3 weeks of waiting for a functional GUI? Built in just 14 minutes [read]
Try the Latest Updates in Refact.ai
Update your VSCode or JetBrains plugin to access the latest features. Vibe coding is the future of software development — get it today with Refact.ai Agent!