Terminal Command Generation, Virtual Quota Provider & Inline Assist Evolution

Features & fixes in Kilo Code v4.61.0-v4.79.0

Welcome back to another Kilo Code digest! After two weeks of hard work, we're back with a mega-update that's thicker than your morning coffee. Terminal commands getting you down? Well, now you can just describe what you want in plain English. Who can remember whether it's grep -r or find -name when you're trying to hunt down that pesky bug at 3 AM?

TL;DR Summary

The highlights you can't afford to miss:

• 🚀 Terminal Command Generator - Generate terminal commands from natural language with Ctrl+Shift+G

• 💰 Virtual Quota Fallback Provider - Set cost/request limits and automatically switch between AI providers

• ⚡ Inline Assist Evolution - Auto-trigger suggestions, better JSON parsing, and comprehensive context awareness

• 🤖 New Models: GPT-5 and Opus 4.1 - The latest and most powerful models are here

• 🎯 Morph Fast Apply - Experimental support for faster and more precise code edits

• 📊 Cost Tracking Improvements - Per-request cost visibility for Kilo Code provider

• 🌍 New AI Providers - Hugging Face, SambaNova, Doubao, and expanded model support

Noteworthy Changes

Terminal Command Generator

Available in: v4.71.0 | Contributor: @hassoncs

Finally, a feature for those of us who treat the terminal like that one relative at family gatherings - we know we should interact with it, but we're never quite sure what to say. Press Ctrl+Shift+G (or Cmd+Shift+G on Mac) and describe what you want in plain English. Want to "find all JavaScript files modified in the last week"? Just type that. Need to "kill all node processes"? Say it like you mean it.

The generator:

• Works in any terminal within VS Code

• Shows helpful tips when you open new terminals

• Let you choose which AI provider to use in settings

Navigate to Settings → Terminal to configure your preferred AI provider for command generation.

Virtual Quota Fallback Provider

Available in: v4.65.0 | Contributor: @mcowger COMMUNITY

Ever blown through your OpenAI credits faster than a developer burns through snacks during a hackathon? The new Virtual Quota Fallback Provider has your back. Set cost or request limits for your profiles, and Kilo Code will automatically switch to backup providers when you hit your limits.

Perfect for:

• Budget-conscious developers who still want premium models when it matters

• Teams sharing API keys with spending limits

• Testing expensive models without the anxiety

Short Instruction (Test it out and let us know what you think! )

• Enable the Virtual Provider experiments flag in settings

• Create at least 3 separate, non-virtual provider profiles

• Create a new provider profile and set its "API Provider" to "Virtual".

• In the Virtual provider settings:

• Select a profile as the Primary Provider.

• Select your second profile as the Secondary Provider.

• Select your third profile as the Backup Provider.

• Set the Virtual provider as your active profile.

(You can test it by setting the first provider to a very low rate limit, e.g., 1 Request per minute.)

More details in the Pull Request

Inline Assist Evolution

Available in: v4.76.0, v4.77.0, v4.67.0 | Contributors: @catrielmuller

Inline Assist got smarter than your average bear. Auto-trigger now shows code suggestions after a configurable delay - it's like having a very eager junior developer looking over your shoulder, except this one actually knows what they're doing.

Key improvements:

• Auto-trigger suggestions after configurable delay

• Better JSON parsing compatibility across more models

• Comprehensive context awareness for more relevant suggestions

• Dedicated settings page for provider selection

• Improved preview styles for suggestion display

• Toggles to disable features when you need focus time

GPT-5 and Opus 4.1 Support

Available in: v4.79.0

The newest and most powerful models have landed, bringing enhanced reasoning capabilities and better code understanding. It's like upgrading from a car to a rocket ship, except the rocket ship can also explain quantum physics while debugging your TypeScript.

Be aware: it’s fancy but also costly. At the moment of this post's publication, the Opus 4.1 output tokens were a whopping $75 per million!

Morph Fast Apply

Available in: v4.71.0 | Contributor: @bhaktatejas922

Experimental support for Morph Fast Apply brings speedier and more precise code edits. It's in experimental mode, so proceed with caution, but the rewards might be worth it!

Cost Tracking Improvements

Available in: v4.61.0, v4.64.0 | Contributor: @chrarnoldus

The Task view now shows per-request costs when using the Kilo Code provider. Visual indicators also appear when cost retrieval fails, because ignorance isn't always bliss.

Developer Experience Improvements

Enhanced Memory Management

Available in: v4.70.2, v4.79.0 | Contributors: @catrielmuller, @chrarnoldus, @xyOz-dev

• Optimized chat view memory usage

• Fixed memory leak in virtual scrolling implementation

• Improved webview memory allocation

Terminal Experience

Available in: v4.71.0, v4.74.0 | Contributors: @chrarnoldus, @hassoncs

• Disabled terminal shell integration by default for cleaner experience

• Terminal generation tip now shows only once per session

UI/UX Refinements

Available in: v4.78.0, v4.71.0 | Contributors: @ABODFTW, @hassoncs, @catrielmuller

• Task timeline now scrolls horizontally with mouse wheel

• Simplified Welcome/Login screen design

• Fixed Profile selector padding issues

• Fixed rule toggle UI functionality

Model Support Expansions

Available in: Multiple versions | Various Contributors

• GPT-5 Support for OpenAI and OpenRouter providers (v4.78.0, v4.79.0)

• GLM-4.5 Series including Flash variant (v4.74.0, v4.75.0)

• Gemma 3 27B for Gemini Provider (v4.75.0)

• Cerebras Updates with latest models including Qwen 3 Coder (v4.72.0, v4.77.0, v4.79.0)

• Z.AI & BigModel providers for GLM-4.5 Series (v4.74.0)

Major Roo Code Integrations

Source: Merged from Roo Code v3.23.14, v3.23.19, v3.24.0, v3.25.4, and v3.25.8

We've integrated five major Roo Code releases, bringing over 100+ improvements upstream. Here's what landed:

New Provider Integrations

• Hugging Face - Full support for open source models

• SambaNova - Enterprise-grade AI infrastructure

• Doubao - ByteDance's AI platform

• Moonshot - Additional provider options

• Mistral Embeddings - Enhanced embedding capabilities

Enhanced Developer Experience

• Code Indexing for Multiple Folders - Similar to task history, now works across your entire workspace

• Message Queueing System - Smoother handling of multiple requests

• Agent Rules Standard (AGENTS.md) - Define custom agent behaviors and rules

• Redesigned Task Header & History - Cleaner, more intuitive interface

• Search in Mode Selector - Quickly find the mode you need

• Terminal Command Permissions UI - Control what commands can be executed

• Markdown Table Rendering - Finally, tables that look like tables

Stability & Performance

• Memory Leak Fixes - ChatView virtual scrolling and DirectoryScanner improvements

• Better Error Recovery - Qdrant service availability detection

• Smarter Command Validation - Improved handling of chained commands and subshells

• Enhanced Token Counting - Using VSCode LM API for accuracy

• Global Rate Limiting - For OpenAI-compatible embeddings

• Configurable Timeouts - Command execution and diagnostic delays

Tool & Feature Improvements

• Image Support in read_file - Process images alongside text

• Auto-Approved Cost Limits - Set spending boundaries that enforce themselves

• Cloud Sync for Provider Profiles - Share configurations across devices

• Dynamic Tool Selection - Architect mode intelligently picks the right tools

• Better AGENTS.md Symlink Support - Works with your existing file structure

• Task Mode Persistence - Modes stick to their tasks

Platform Enhancements

• Claude Code Windows Support - Native Windows compatibility

• Improved Vertex AI Integration - Better model name formatting

• Bedrock API Key Support - AWS integration improvements

• Organization-Level MCP Controls - Team-wide configuration management

• ESC Key Navigation - Close popovers naturally

Performance & Stability Fixes

Provider & Model Fixes

• Ollama: Correct context window size reporting and usage (v4.66.0)

• Ollama: Increased timeouts from 5 minutes to 1 hour (v4.74.0, v4.72.0)

• LM Studio: Fixed model context length issues (v4.79.0)

• Fireworks: Provider fixes (v4.75.0)

• KiloCode: Fixed "Failed to load provider model list" errors (v4.79.0)

Code Application & Processing

• VB.NET: Fixed indexing with fallback chunking system (v4.79.0)

• Swift: Added files to fallback list (v4.79.0)

• Git Diff Views: Fixed 'failure to apply changes' when views are open (v4.68.0)

• Message Queue: General fixes and improvements (v4.75.0)

UI & Interaction Fixes

• Fixed Kilo Code inadvertently stealing focus (v4.61.0)

• Fixed missing Kilo icon bug by switching to PNG (v4.68.0)

• Fixed UI for approving chained commands (v4.79.0)

• Improved highlight layer style alignment with textarea (v4.79.0)

Polish & Bug Fixes

Localization & Accessibility

• Traditional Chinese translation refinements (v4.72.0, v4.73.1)

• Standardized brand names in localizations (v4.72.0)

• Removed low-usage language support (Filipino, Greek, Swedish) (v4.63.0)

• Fixed linter not applied to locales/*/README.md (v4.79.0)

Settings & Configuration

• Claude Code system prompt length increased (v4.75.0)

• Thinking Budget value parsing and boundary handling corrected (v4.79.0)

• Added "Generate terminal command" label on prompts settings page (v4.74.0)

• Fixed Orchestrator mode sub-task result reporting (v4.74.0)

• Fixed MCP Marketplace installation modal state issues (v4.74.0)

Small But Important

• Backend notifications support added (v4.68.0)

• Credits can be purchased directly from profile tab (v4.70.2)

• Kilo Code icon now adapts to light/dark themes (v4.62.0)

• Project ID loading from Gemini CLI's .env file (v4.63.0)

• Kilo Code provider logout on data reset (v4.61.0)

• OpenRouter provider routing strategy selection (v4.63.0)

Get the Latest Release

Ready to supercharge your coding with these updates? Download the latest Kilo Code release: Download v4.79.0

Follow us for more updates and join the conversation about the future of AI-powered coding in our Discord chat!

---

Remember: Terminal commands are like spells - with great power comes great responsibility. And now with the Terminal Command Generator, you don't even need to memorize the incantations. Just don't ask it to rm -rf / - it might actually listen.