Text classification models

Metatext.AI hosts prebuilt models that read text and return a label or category. A Beam agent sends incoming text to a model, reads the prediction, and applies your rule for what each label should do next, such as tagging a record or routing a message. Confident predictions within your accepted range are written back automatically. Low-confidence results, or inputs the model was not built to handle, are set aside for a person to label by hand. The agent uses model output to act, while people resolve the cases the model is unsure about.

Prediction write-back

Once a Metatext.AI model returns a result, it needs to land where your systems can use it. A Beam agent takes the prediction, reads the record it belongs to, and writes the label or score back to the field your workflow reads from. Records that update cleanly proceed without review. Predictions that conflict with an existing value, or that fall below your confidence floor, are flagged to a person before anything is overwritten. The agent keeps the write path consistent, and a human settles cases where the model and the record disagree.

Model management

Metatext.AI lets you create and maintain the models that read and write text for your team. A Beam agent works with the models you have already set up, reads their configuration, and applies your rules for which model handles which input. Requests that fit an existing model are handled and recorded automatically. When a request needs a model that does not exist, or a configuration change to an existing one, the agent stops and hands the decision to a person, since setting up or altering a model stays a human responsibility.