Manual Agent Setup

Agent Basics

Give your agent a name and describe what it will do.

Upload Knowledge Base

Upload documents your agent will reference when answering questions.

Training Setup

Define what your agent learns to decide about each item in your documents, then adjust how much emphasis each file gets during training.

Classification tasks

Your agent learns by labeling every item in your documents. An item is one unit of your data: one line or row for statements and spreadsheets, one message for emails and tickets, one document for contracts and reports.

A classification task is one question the agent answers about each item, with a fixed menu of answers. Name the question, then list every allowed answer as a label.

Examples: task expense_category with labels Meals / Travel / Office / Other. Task urgency with labels High / Medium / Low. Task action_needed with labels File / Review / Escalate.

Tips: labels should cover every item you expect to see, so include a catch-all like "Other" or "Ignore". Aim for 2 to 10 labels per task. After you submit, a teacher model labels a sample of your real items first and you review its answers before any training starts, so you will see exactly how your tasks behave on your own data.

+ Add Task

Data weights

Connect Integrations

Choose up to 3 APIs to connect to your agent. Select capabilities for each.

Orchestrator Model

Choose the AI model that will power your agent's reasoning and analysis.

Review & Create

Review your agent configuration before creating.