Open source AI assistant OpenClaw shipped version 2026.4.26 with three big changes: a new one-click migration tool, full QQ group chat support for its bot, and a major overhaul of its Ollama local model integration.
The new openclaw migrate command can import configs, memory, model providers, MCP servers, skills, and credentials from Claude Code/Desktop and Hermes in one shot. It automatically backs up before migrating and supports preview and dry runs.
On the QQ bot front, group chats now get @mention wake-up, per-group independent configs, and message queue sorting; private chats get streaming messages; large file sending is switched to chunked upload.
The Ollama integration got the broadest coverage. OpenClaw can now auto-read context length and tool support parameters from Ollama model configs — no more manual duplication. Thinking effort support now matches Groq and LM Studio's native level. On networking, remote Ollama endpoints, LAN/Tailnet internal addresses can be directly configured and trusted. Embedding switched to batch API for better performance.
Also new: Tencent Yuanbao and Cerebras channels; Matrix gets one-click end-to-end encryption; Google Meet integration supports joining meetings directly via OpenClaw browser with auto audio/video permission acquisition. On the ops side, overly long conversation logs get auto-compressed, scheduled tasks get concurrency control and delivery result classification, and the config edit panel gets change comparison and sensitive value masking. Over 200 bug fixes — a number that makes sense given reports that OpenClaw's rapid update pace was causing stability headaches for big companies like Nvidia and Tencent.