1. More AI Brains to Choose From

Your AI assistant can now connect to even more models:

Venice AI: A new AI option added to the lineup
LM Studio: Run your favorite models locally — connection is now smoother
MIMO: Another new model provider
Azure OpenAI: Better support for enterprise users on Microsoft’s cloud

More choices mean you can pick the AI that works best for you

2. More Platforms to Chat On

Your AI assistant can now reach you in more places:

VK: Finally available for Russian users — AI comes to VK channels
Feishu (Lark): Minor fixes, emoji reactions are more stable
DingTalk: Group @ mentions are more accurate — no missed or duplicate replies
Telegram: Quoted replies now include context and media
WeChat: Several bug fixes, conversation context now persists across restarts

One line: AI can chat with you in more places, and the experience is smoother

3. Toolkit Upgrades

The tools your AI uses to get things done are more powerful:

Web search: You can now specify a time range, e.g. “only results from the past week”
File reading: Supports reading by line range, more flexible for large files
Load image: New load_image tool lets AI process local images
Background exec: exec tool now supports background execution and PTY — long tasks won’t block
Reaction tool: AI can now react to messages with emoji

4. Web UI Refresh

Open the web UI and you’ll notice:

🎉 First-time tour: A guided walkthrough for new users — no more guessing

First-time tour

🖼️ Image messages: The web chat now displays image messages

Image messages

🔧 Log level controls: Adjust log verbosity right from the web UI — easier debugging

Log level controls

🛒 Skill Marketplace: Browse and install extension skills with one click

Skill Marketplace

🔐 Dashboard login protection: Launcher dashboard is now protected with a token

🏷️ Version display: Backend version shown in the config page header — easy to confirm

5. Under the Hood

You won’t see these directly, but they matter:

🚦 LLM rate limiting: Requests queue up automatically instead of crashing
🧠 Context manager: New ContextManager abstraction for smarter long-conversation handling
📊 Token estimation fix: Fixed double-counting of tokens — no more false overflow warnings
🔄 Auto-updater rewrite: Version selection and extraction are now more reliable
📝 Log file support: PICOCLAW_LOG_FILE env var lets you write logs to file only

6. Bug Fixes

✅ GitHub Copilot session creation failed — fixed
✅ WeChat context lost after restart — fixed (now persisted to disk)
✅ Gateway reload caused Pico to stop working — fixed
✅ Startup crash (FlexibleStringSlice issue) — fixed
✅ Log panel scroll behavior — fixed
✅ Skills page dark mode colors — fixed
✅ Discord token changes not saving — fixed
And dozens more small fixes

In One Line

v0.2.5 makes PicoClaw more capable (new AI models + new tools), easier to use (refreshed UI + first-time tour), and more stable (dozens of bug fixes) — a solid upgrade

PicoClaw — Lightweight, cross-platform, fast

Website: picoclaw.io

GitHub: github.com/sipeed/picoclaw

Docs: docs.picoclaw.io

Discord: discord.gg/V4sAZ9XWpN