Fara-7B: A Tiny AI Agent That Runs Your Web Chores Privately

Fara-7B is a new open-weight computer use agent that understands screenshots and text to complete multi-step web tasks. It takes a high-level goal like “book a restaurant” and plans and executes the needed actions, clicking buttons, typing, and navigating pages on its own. With just 7 billion parameters, this model is small enough to run locally on consumer-grade GPUs.

Microsoft Research developed Fara-7B to tackle the lack of high-quality training data for computer use agents. The team created FaraGen, a synthetic data pipeline that generates verified web task trajectories at scale, each costing about a dollar. They trained the model on these trajectories and released it under an MIT license, making it freely available for anyone to use or modify.

Efficient on-device computer use

Privacy-conscious professionals, small agencies, and serious hobbyists can benefit from running this model on their own hardware. Local execution keeps sensitive data off cloud servers, reduces latency, and avoids API costs. It automates everyday web chores like shopping, travel booking, and information gathering while respecting safety stops that require human approval for purchases or personal data entry.

Safety and development notes

The model refuses illegal, deceptive, or high-risk tasks and pauses at points where personal information or payment is needed. Microsoft’s red-teaming assessed risks such as jailbreaks and harmful content, and they recommend sandboxed environments and human-in-the-loop oversight. Currently English-only, future versions may expand language support and refine the synthetic data pipeline further.

“Fara-7B is an ultra-compact Computer Use Agent (CUA) that achieves state-of-the-art performance within its size class and is competitive with larger, more resource-intensive agentic systems.” — Source: Hugging Face