Learning Timeline
Key Insights
Choosing the Right Model
Models like Qwen or Llama 2 (base versions) might be too limited for complex web navigation. Use DeepSeek-R1 14B or higher for more consistent results.
Advantages of Local LLM vs Cloud
Local LLMs have the potential to solve 'Captchas' that are typically blocked by cloud-based AI like ChatGPT Operator due to security policy restrictions.
Automatic Attempt Limits
The system includes a self-healing feature where it will retry up to 5 times if it fails to interact with a web element before stopping completely.
Prompts
Automated E-Commerce Task
Target:
DeepSeek-R1 via Ollama
Go out to network Chuck coffee, find the 404 error coffee and add it to my cart.
Step by Step
How to Configure a Local LLM Provider
- Click on the 'LLM configuration' option found in the sidebar or main menu.
- Under the 'LLM provider' section, click the dropdown menu and select 'Ollama'.
- Go to the 'model name' section to choose the AI model you want to use.
- Type or select the specific model name (e.g., 'deepseek-r1:14b').
- Ensure the model has been downloaded via the Terminal before proceeding (refer to the Snippet Card).
Running Your First Autonomous Task
- Click on the 'Run Agent' tab within the Web UI interface.
- Enter the instruction or task you want the AI to perform in the input field (or select the 'demo option' for a quick trial).
- Click the 'Run Agent' button to start the process.
- Monitor the browser window that opens automatically; the AI will begin labeling and numbering elements on the webpage for navigation.
- Let the AI work; if it fails to perform an action, the system will trigger an 'Autocorrecting' process automatically for up to 5 attempts.
