Kilo Code Weekly Product Roundup | March 4, 2026
Review Mode Suggestions, Claude Sonnet 4.6, New Providers, and 30+ Community PRs
Welcome back to the weekly product roundup! This week brings one-click review suggestions after task completion, Claude Sonnet 4.6 across three providers, two new API providers, GLM 5 support, and an enormous wave of community contributions.
VS Code Extension
One-Click Review Suggestions
After completing a task in Code or Orchestrator mode, Kilo now suggests starting a review. You’ll see options to jump into Review mode, or switch directly to Code, Debug, Architect, or Orchestrator based on the review findings. There’s also an option to clear context and start a fresh review of uncommitted changes. It’s a small prompt that closes the loop between building and reviewing without breaking your flow.
Claude Sonnet 4.6 Support
Claude Sonnet 4.6 is now available across Anthropic, Bedrock, and Vertex providers. Select it from your model picker and you’re good to go.
GLM 5 and Z.ai Updates
GLM 5 is now supported, with Z.ai defaulting to glm-5. Model selection is also now consistent between VS Code and webview settings.
Zenmux Provider
Zenmux is now available as a new provider, along with fixes to its context window detection (preventing erroneous context-condensing loops) and improved tool-calling reliability.
New Models
MiniMax M2.5, M2.5-highspeed, and M2.1-highspeed added
Gemini 3.1 Pro default metadata updated with consistent tool calling behavior
Deprecated Cerebras models removed (llama-3.3-70b, qwen-3-32b)
Stability & Fixes
Kill Command reliability: The “Kill Command” button now reliably terminates processes on all platforms, including background processes
Image preservation on edits: Attached images are no longer silently dropped when editing a message with a checkpoint
Cost display: Better formatting for low cost values so you can actually read sub-cent amounts
Agent Manager scroll: Fixed scroll jump when reading long completion messages
OpenAI reasoning content: Reasoning content is now preserved in OpenAI format conversion
Anthropic extended thinking: Enhanced compatibility
API cost persistence: Total API cost now persists correctly after message deletion
Skills notifications: Visual bug fixes in Agent Behaviour settings page
KiloClaw
Dashboard Improvements
The KiloClaw Dashboard’s “Redeploy” section now clearly presents two options: Redeploy (keep the same platform version) or Upgrade & Redeploy (upgrade to the latest supported version). The “when to redeploy” guidance now points users toward Upgrade & Redeploy for changelog updates and picking up the latest platform version. A new Restore Config button in Settings lets you restore the original openclaw.json, with automatic backup to /root/.openclaw.
Code Reviewer
Roast Mode
Set Code Reviewer to “Roast” mode to hear what the AI model really thinks of your code - unfiltered, and brutally honest. It’ll still catch your issues, but it won’t be nice about it…
Install & Update CLI
npm install -g @kilocode/cli
Community Contributions
A huge couple of weeks for the community. Thank you to everyone who contributed:
#5247 - @theQuert - Apertis API provider
#5883 - @Astricaelus - MiniMax M2.5 and highspeed models
#5526 - @nubol23 - Voyage AI embedder support
#5878 - @CoinAnole - Moonshot coding endpoint fix
#5746 - @Githubguy132010 - Dev container persistence
#5575 - @Patel230 - maxReadFileLine=0 fix
#5639 - @mikij - Agent Behaviour settings visual fixes
#5904 - @Githubguy132010 - Agent Manager scroll fix
#5811 - @wombatepiclandingstudio - Context token indicator fix
#5565 - @Githubguy132010 - File deletion checkbox fix
#5377 - @benzntech - MODEL_NO_TOOLS_USED error loop fix
#5586 - @abdulrahimpds - Settings search bar UI fix
#5760 - @Githubguy132010 - User message theme-aware colors
#5752 - @Madrawn - Slash commands in tool_result blocks
#5055 - @Leoyzen - Preserve reasoning content in OpenAI format
#5739 - @rayss868 - Honor explicit reasoning effort disable
#5167 - @hdcodedev - Kill Command reliability fix
#5630 - @SenatusSPQR1 - NanoGPT reasoning extraction
#5814 - @shssoichiro - Custom Ollama embed dimensions
#5523 - @abdulrahimpds - Persist API cost after message deletion
#5897 - @evanjacobson - Prevent MCP server restart on settings re-read
#5150 - @hdcodedev - Duplicate tool_use/tool_result ID fix
#5896 - @evanjacobson - Provider settings profile switching fix
#5149 - @dacsang97 - Anthropic extended thinking compatibility
#5816 - @evanjacobson - Numeric tool call ID fix
#5576 - @Patel230 - Symlink handling in skills directory
#5900 - @evanjacobson - Abort listener memory leak fix
#5459 - @Schrolli91 - Low cost value formatting
#5831 - @Neonsy - ZenMux context window and tool-calling fixes
#6017 - @PeterDaveHello - Gemini 3.1 Pro metadata update
#5901 - @SkipperQ93 - JetBrains editor initialization fix
#4768 - @hsp-sz - Zenmux provider
#4714 - @otterDeveloper - Fireworks.ai model updates
#4926 - @YuriNachos - Disable zsh history expansion fix
#5162 - @hdcodedev - Image preservation on message edit
#5139 - @naga-k - Prevent thinkingLevel for unsupporting Gemini models
#4945 - @CaiDingxian - Chars count in ListFilesTool
#5805 - @Neonsy - GLM 5 support and Z.ai defaults