Local model inference

Ollama runs large language models on your own hardware. A Beam agent sends a prompt to a chosen local model when a step needs generated text, reads the response, and applies your rules to use it, such as drafting a reply or summarizing a document. It writes the output into the workflow where it belongs. Responses that fall below a confidence bar you set, contain flagged content, or concern sensitive topics route to a person for review before use. Running inference locally keeps prompts and data inside your environment rather than a third party.

Document summarization

A local Ollama model can condense long text into short summaries. A Beam agent feeds documents or threads to the model, reads the summary, and writes it onto the related record so people skim instead of reading in full. It applies your rules for length and focus. Summaries of legal, financial, or otherwise sensitive material, or ones the agent judges uncertain, route to a person who checks accuracy before the summary is relied upon. The agent produces the routine summaries you approved and never lets an unverified one drive a consequential decision alone.

Text classification and tagging

An Ollama model can sort text into categories you define, such as topic, sentiment, or intent. A Beam agent runs incoming messages or tickets through the model, reads the predicted label, and applies your rules to tag and route each item. It writes the label to the record so downstream steps act on it. Items the model labels with low confidence, or that fall between categories, route to a person who assigns the correct tag. The agent classifies the clear cases automatically and keeps ambiguous ones under human eyes.