# AI use cases for Fast Browser Use (APUS-AI-Lab/fast-browser-use) Repo: [APUS-AI-Lab/fast-browser-use](https://github.com/APUS-AI-Lab/fast-browser-use) | Use cases: 3 | Page: https://dedupedai.com/use-cases/apus-ai-lab-fast-browser-use/ Fast Browser Use is a local-first browser automation engine and agent skill that uses open Qwen3.5 models to pick page actions from visible elements. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Check a web settings form saves correctly, offline: https://dedupedai.com/use-cases/apus-ai-lab-fast-browser-use/verify-web-form-changes-locally/ - Give a coding agent a fast local browser for page tasks: https://dedupedai.com/use-cases/apus-ai-lab-fast-browser-use/let-coding-agent-drive-browser-offline/ - Measure task time for two local Qwen models on your sites: https://dedupedai.com/use-cases/apus-ai-lab-fast-browser-use/benchmark-local-models-on-browser-tasks/ ## How to check a web settings form saves correctly, offline with Fast Browser Use Repo: [APUS-AI-Lab/fast-browser-use](https://github.com/APUS-AI-Lab/fast-browser-use) | Difficulty: Advanced | Time: about 90 minutes | Basis: From the README Fill and save a settings form in a local browser run, then confirm the saved message and page state. **Who it is for:** A developer or QA engineer who must confirm a web app settings page saves correctly and wants the check to run on their own machine without cloud calls. **The problem:** Scripted browser tests break when selectors change, and cloud browser agents send page content off the machine. **The outcome:** A browser run that fills, selects, toggles and saves the form, then reports an execution trace that you can check against the saved confirmation. ### Steps 1. Clone the repo with `git clone https://github.com/APUS-AI-Lab/fast-browser-use` (inferred, check the README) 2. Install the prerequisites listed in the README Quickstart: Python 3.12+ and uv. 3. Download the Qwen3.5-9B weights and set up the engine as described in the README Quickstart section. 4. Follow the Agent Skill Setup section if you want Claude Code, Codex or Cursor to call the engine. 5. After the run, check the page with external assertions such as `--expect-url`, `--expect-title` or `--expect-text` rather than trusting the model's DONE output. ### Prompt to paste into your agent ```text Use the fast-browser-use repo at to fill and save the settings form at
with the values in . Return the execution trace and whether the saved confirmation text matches . Verify the page with the --expect-text assertion before reporting success. ``` ### Caveat The README excerpt does not show the full install commands, so the setup steps depend on the Quickstart and Agent Skill Setup sections in the complete README. Risk, credentials: If the form sits behind a login, the agent will type text into fields, so keep account credentials out of shared prompts and logs. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/apus-ai-lab-fast-browser-use/verify-web-form-changes-locally/ ## How to give a coding agent a fast local browser for page tasks with Fast Browser Use Repo: [APUS-AI-Lab/fast-browser-use](https://github.com/APUS-AI-Lab/fast-browser-use) | Difficulty: Intermediate | Time: about 60 minutes | Basis: From the README Let Claude Code, Codex or Cursor handle planning while a local engine handles clicks, selects and scrolls. **Who it is for:** A developer using an AI coding agent who needs it to navigate a website or docs page during a task and wants low latency with no cloud inference. **The problem:** Cloud browser agents add seconds per step and send full pages or screenshots to a remote service. **The outcome:** The host agent plans the task and calls the local engine for page actions, then checks the final page state against ground truth. ### Steps 1. Clone the repo with `git clone https://github.com/APUS-AI-Lab/fast-browser-use` (inferred, check the README) 2. Install Python 3.12+ and uv, then set up the local Qwen3.5 model as described in the README Quickstart. 3. Install the skill for your host by following the Agent Skill Setup section, which covers Claude Code and Codex. 4. Ask the host agent to hand page steps to the engine and keep planning and code analysis in the host. 5. Review the returned execution trace and confirm the final URL or title with an external check. ### Prompt to paste into your agent ```text Using the fast-browser-use skill in , complete the browser task on . Return the step-by-step execution trace and the final page URL and title. Verify the final URL matches before marking the task done. ``` ### Caveat The README names Claude Code, Codex and Cursor as hosts, but the exact skill install commands are not in the excerpt, so verify them in the full README. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/apus-ai-lab-fast-browser-use/let-coding-agent-drive-browser-offline/ ## How to measure task time for two local Qwen models on your sites with Fast Browser Use Repo: [APUS-AI-Lab/fast-browser-use](https://github.com/APUS-AI-Lab/fast-browser-use) | Difficulty: Advanced | Time: about 120 minutes | Basis: From the README Run the same browser task on Qwen3.5-9B and Qwen3.5-35B-A3B and compare task times and verification results. **Who it is for:** A researcher or engineer choosing between a smaller and a larger local model for browser automation on their own hardware. **The problem:** Published numbers come from one hardware setup, so you cannot tell how a model performs on your machine and tasks. **The outcome:** A set of task times and pass or fail checks for both models that you can compare with the README benchmark tables. ### Steps 1. Clone the repo with `git clone https://github.com/APUS-AI-Lab/fast-browser-use` (inferred, check the README) 2. Install Python 3.12+ and uv, then download both Qwen3.5-9B and Qwen3.5-35B-A3B weights as described in the README Quickstart. 3. Pick two or three browser tasks, such as the Wikipedia navigation task, and write the strict URL or title check for each. 4. Run each task three times per model and record the task time, as the README benchmark does with three trials. 5. Compare your median times with the published tables in docs/performance.md and the telemetry in docs/qwen9b-demo-measurement.json. ### Prompt to paste into your agent ```text In the fast-browser-use repo at , run the task three times each on Qwen3.5-9B and Qwen3.5-35B-A3B. Return a table of task times and whether the final URL and title matched . Verify that every run used local inference only before reporting results. ``` ### Caveat The published benchmarks come from one NVIDIA RTX PRO 6000 Blackwell workstation, so your timings will differ on other hardware. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/apus-ai-lab-fast-browser-use/benchmark-local-models-on-browser-tasks/