Workers are AI models that carry out tasks within Pulse based on your custom-defined ACL (Access Control List) rules. Each worker interprets these rules and performs tasks such as summarization, search processing, entity extraction, or content transformation.
š§ How Workers Function
- Workers are invoked by ACL rules to respond to content triggers, queries, or ingest events.
- Each worker is tied to a specific model (e.g., Qwen, GPT-4o-mini) and executes tasks appropriate to its model's capabilities.
- Tasks are evaluated and completed locally by the model; heavier operations should be routed to more capable models when available.
š Notes on Usage
- š§āš» Non-commercial accounts currently have access to Qwen only (
qwen-2.5-vl-7b-instruct). - āļø Choose your model wisely: A 7B model like Qwen is efficient and fast, but not suitable for complex reasoning or large-scale analysis. Models like GPT-4o or Claude should be used for higher-cognition tasks.
- š§© Model size and latency affect outcome quality. Ensure the task assigned to a worker is within that model's feasible range to avoid partial or low-quality responses.
ā Best Practices
- Keep worker roles focused: one model for summaries, another for entity extraction, etc.
- Define task thresholds clearly in ACL rules to prevent small models from being overloaded.
- Use the "Edit" interface to assign model type, description, and rules in a modular, maintainable way.