Tether's New BotPsy AI Agent Beats Larger Models on Phone Tasks

Tether has introduced BotPsy, a compact AI agent featuring 800 million parameters designed to carry out everyday tasks on personal devices, according to Decrypt. In company testing, the relatively small model outperformed competitors up to 2.5 times its size on five of six tool-use benchmarks.

Unlike traditional chatbots, BotPsy is agentic and built to take action by identifying required tools, executing commands in sequence, and adjusting when operations fail. Built on Alibaba's Qwen3.5-0.8B architecture, the model scored 44.0% on the BFCL v4 benchmark and 58.7% on ACEBench, surpassing both its base model and larger variants.

Local Execution and QVAC Ecosystem

BotPsy expands Tether's QVAC ecosystem, which aims to provide local AI models running on user-controlled devices rather than centralized cloud infrastructure. The ecosystem previously introduced models for medical applications, translation, and speech processing, including the QVAC MedPsy system released earlier this year.

By running locally, models like BotPsy can manage multi-step tasks involving phone settings, messaging, and scheduling without transmitting personal information to cloud servers. The smaller parameter count reduces the memory required for deployment on consumer hardware.

Availability

BotPsy is currently available as an open-weight model under the Apache 2.0 license, allowing developers to download, modify, and integrate it into their own applications. Tether has also released compressed versions to further reduce memory requirements for local deployment.

Key facts

• BotPsy features 800 million parameters and is built on Alibaba's Qwen3.5-0.8B architecture.
• The model outperformed competitors 2.5 times its size on five of six tool-use benchmarks in company testing.
• BotPsy scored 44.0% on BFCL v4 and 58.7% on ACEBench.
• The model is available as an open-weight release under the Apache 2.0 license within Tether's QVAC ecosystem.