Kilo Code Weekly Product Roundup | Feb. 18 2025
Welcome back to the weekly product roundup! This week brings Roast Mode for Code Reviews, two new hosted platform features (Gas Town and KiloClaw), Claude Sonnet 4.6 support, a wave of new models and providers, one-click fix suggestions in Review mode, and a solid round of stability and reliability improvements across the extension.
Code Reviewer — Roast Mode
Code Reviewer Roast Mode gives you brutal, unfiltered PR feedback — the kind a very blunt senior engineer might offer at 4pm on a Friday. Think “You named this variable temp2. Temp for what? Temporary sanity?” or “I see you copied this from Stack Overflow. The answer had 3 downvotes.”
If you want the comical truth about your code, try it now at app.kilo.ai/code-reviews and select Roast Mode.
Kilo x Gas Town
Gas Town is an agent orchestrator by Steve Yegge that coordinates dozens of AI coding agents simultaneously — think Kubernetes, but for coding agents. Kilo will soon be hosting it on managed cloud infrastructure with native Kilo Gateway integration.
What that gets you: 500+ AI models, unified billing, automatic updates, monitoring, and zero ops overhead. No tmux tunnels, no Mac Minis in closets, no 3am crash pages. Spin up your Mayor, Deacon, Witness, and Refinery in minutes, and scale your polecat swarms up or down on demand.
Self-hosting Gas Town works, but you’re signing up to own every failure mode — persistent compute, key management across providers, and being your own on-call when things break overnight. Kilo x Gas Town handles all of that.
KiloClaw
OpenClaw (209k+ GitHub stars) is the fastest-growing open source AI agent project right now. It connects to 50+ chat platforms, runs shell commands, controls browsers, manages files, and remembers everything across sessions. KiloClaw hosts it for you.
One-click deploy: go from zero to running OpenClaw in under 60 seconds. No SSH, no Docker, no YAML. Your Kilo Code account is your KiloClaw account — same credits, same billing, same 500+ models through Kilo Gateway.
VS Code Extension
Review Mode: One-Click Fix Suggestions
Review mode now surfaces one-click suggestions to apply fixes directly from the review. Based on what the reviewer finds, it’ll recommend switching to Code, Debug, Architect, or Orchestrator mode — so you go from “here’s the problem” to “fix it” without any extra navigation.
Claude Sonnet 4.6
Claude Sonnet 4.6 is now supported across the Anthropic, Bedrock, and Vertex providers.
New Models and Providers
Apertis added as a new API provider
MiniMax-M2.5, MiniMax-M2.5-highspeed, and MiniMax-M2.1-highspeed models added
Voyage AI embedder support added
GLM 5 support added for Z.ai, with
glm-5set as the new defaultMiniMax 2.1 and GLM 4.7 added to Fireworks.ai, with other model updates
Zenmux added as a new provider
Custom embed dimensions now supported for the Ollama provider
Quality of Life
Agent Manager scroll fix: no more scroll jumps when reading long completion messages
User message visibility: improved contrast using distinctive theme-aware colors
Low cost values: better formatting for sub-cent token cost displays
Dev container persistence: threads and settings now persist in dev containers
Chars count added to the ListFiles tool
Reliability and Bug Fixes
MCP server restarts: prevented MCP servers from restarting on every settings file re-read
Numeric tool call IDs: fixed tool use failures for providers (like MiniMax) that return numeric IDs by coercing to string in the shared stream parser
Reasoning content preservation: OpenAI format conversion now preserves reasoning content
Reasoning effort: explicit ‘disable’ setting is now properly honored
Anthropic extended thinking: improved compatibility
Total API cost: now persists correctly after message deletion
Gemini thinking level: prevented sending
thinkingLevelto models that don’t support itAttached images on edit: images attached to messages with checkpoints no longer get dropped on edit
Community Contributions
Thanks to everyone who opened PRs and shipped improvements this week:
#5247 — @theQuert (Apertis API provider)
#5883 — @Astricaelus (MiniMax M2.5 and M2.5-highspeed models)
#5526 — @nubol23 (Voyage AI embedder support)
#5878 — @CoinAnole (Moonshot coding endpoint model selection fix)
#5746 — @Githubguy132010 (Dev container persistence for threads and settings)
#5575 — @Patel230 (maxReadFileLine=0 treated as unlimited)
#5639 — @mikij (Agent Behaviour settings visual bug fixes)
#5195 — @Drilmo (Filter internal verification tags from assistant messages)
#5904 — @Githubguy132010 (Agent Manager scroll jump fix)
#5811 — @wombatepiclandingstudio (Context token indicator flicker fix)
#5565 — @Githubguy132010 (File deletion auto-approve checkbox fix)
#5377 — @benzntech (MODEL_NO_TOOLS_USED error loop fix)
#5586 — @abdulrahimpds (Settings search bar UI fixes)
#5760 — @Githubguy132010 (User message theme-aware colors)
#5752 — @Madrawn (Slash commands in tool_result blocks)
#5055 — @Leoyzen (Preserve reasoning content in OpenAI format conversion)
#5739 — @rayss868 (Honor explicit ‘disable’ for reasoning effort)
#5167 — @hdcodedev (Kill Command button reliability fix)\
#5630 — @SenatusSPQR1 (NanoGPT native reasoning field extraction)
#5814 — @shssoichiro (Custom embed dimensions for Ollama)
#5523 — @abdulrahimpds (Persist total API cost after message deletion)
#5926 — @alex-alecu (One-click fix suggestions in Review mode)
#5897 — @evanjacobson (Prevent MCP server restarts on settings re-read)
#5150 — @hdcodedev (Prevent duplicate tool_use/tool_result IDs)
#5896 — @evanjacobson (Fix provider settings leak between profiles)
#5149 — @dacsang97 (Anthropic extended thinking compatibility)
#5816 — @evanjacobson (Fix tool use failure for numeric tool call IDs)
#5576 — @Patel230 (Symlink handling in skills directory)
#5900 — @evanjacobson (Prevent abort listener memory leak)
#5459 — @Schrolli91 (Better formatting for low cost values)
#5831 — @Neonsy (ZenMux context window detection and tool-calling reliability fixes)
#4768 — @hsp-sz (Zenmux provider)
#4714 — @otterDeveloper (MiniMax 2.1, GLM 4.7, and model updates for Fireworks.ai)
#4926 — @YuriNachos (Disable zsh history expansion)
#5162 — @hdcodedev (Fix images lost when editing a message with checkpoint)
#5139 — @naga-k (Prevent thinkingLevel being sent to unsupporting Gemini models)
#4945 — @CaiDingxian (Chars count in ListFiles tool)
#5805 — @Neonsy (GLM 5 support and Z.ai default model update)