Tencent's QQ announced on 15 August that QQ Bot now supports official DeepSeek Harness plugins, turning any QQ bot into an AI agent across both one-to-one and group chats.
The install
Three steps: npm-install the plugin, launch it with npx @deepseek-ai/dsh web, then scan a QR code to bind the account. No manual API key configuration. Reducing agent deployment to a QR scan is the part that matters — it moves the audience from developers who can wire credentials to anyone who can follow three instructions.
What ships with it
Per-user and per-group conversation memory, live model switching mid-chat, an optional mention-only response mode so a bot stays quiet in busy groups, and automatic session recovery after a restart. That last item is unglamorous and load-bearing: agents that lose state on restart are demos, not deployments.
What the news is, precisely
Not the framework. Harness was released on 13 August — outside this window — as a v0.1 developer preview under the MIT licence, built on DeepSeek's Cordis architecture and aimed squarely at Anthropic's Claude Cowork. The news is that Tencent shipped an integration two days later. Nor is Harness production-ready by its own maintainers' account: they warn of compatibility-breaking changes ahead.
The number nobody gave
How many bots are running it. Neither Tencent nor DeepSeek published an install count, and QQ's overall platform scale is not a proxy for Harness adoption. "China's largest chat platform gets AI agents" describes an available capability, not a measured deployment.
Why the speed is the signal
A major platform integrating a two-day-old v0.1 preview from another company is not normal procurement behaviour. It suggests the Chinese agent stack is being assembled in public and at speed, with distribution decisions made before the frameworks underneath them have stabilised — the opposite of the enterprise-first sequencing Western agent products have followed.