Tether AI Research released BotPsy, an open source model with 0.8 billion parameters designed for on-device personal assistants. The system is built on the Qwen3.5-0.8B base and uses a four-stage training pipeline. BotPsy focuses on calling tools for external facts instead of storing general knowledge.
The model achieves state-of-the-art tool-use performance on benchmarks like BFCL v4 and ACEBench. It outperforms larger models such as Qwen3.5-2B-Think on five of six agentic benchmarks. Tether plans to expand BotPsy with decentralized peer-to-peer queries and local agent communication.
The release offers a lightweight option for developers building local AI systems. BotPsy’s performance in lower-resource languages like Hindi and Russian remains a weakness. Developers should test the model thoroughly for multilingual tasks before integrating it into production workflows.
