Agents are the core of Fluent. Each Agent is a configured AI assistant with its own instructions, tools, approval rules, and access settings. Admins create and manage Agents under Manage > Agents.
General tab
Details — the basics:
Name — how the Agent appears to users
Description — a short summary shown on the Agent card
Instructions — the core prompt telling the Agent what to do, how to extract fields, and how to handle edge cases (up to 5,000 characters)
Suggested messages — quick-reply options shown to users to prompt common tasks
Intelligence — model and reasoning settings:
Default model — which AI model the Agent uses (GPT or Gemini variants)
Reasoning effort — how much compute the model applies (Minimal → High). Higher effort is slower but handles ambiguous documents better.
Capabilities tab
Memory — when enabled, the Agent learns from corrections and edits to improve future extractions
Tools — actions the Agent can take, such as fetching ERP data or sending notifications. Each tool can be enabled or disabled and configured individually.
Triggers — automated events that start the Agent without a user initiating the conversation (see Setting up Agent triggers)
Workflow tab
Approvers — team members who must review and sign off on documents before they can be processed (see Setting up document approvers)
Expressions — CEL-based transformations applied to extracted data before it's sent to your ERP; useful for custom field mapping or calculation
Access tab
Sharing — controls which users can interact with the Agent (see Controlling who can use an Agent)
Email — the Agent's inbound email address and sender whitelist for document intake
Save first: Some sections (Approvers, Triggers) only become editable after you save the Agent for the first time.