Walk through the New Agent wizard step by step — naming your agent, choosing a model, writing a system prompt, and running a test before going live.
Open the New Agent wizard
From the dashboard, click the "+" quick-action button or navigate to Agents → New Agent. The wizard opens as a full-screen overlay with a progress bar at the top showing three steps: Configure, Connect, and Test. In the first step you give your agent a name (up to 64 characters) and an optional description. Agent names must be unique within your workspace. Choose a name that describes the agent's purpose — for example, "Customer FAQ Bot" or "Lead Enrichment Agent" — because this name appears in logs, notifications, and team activity feeds.
Choose a model and write a system prompt
In the Configure step, select the underlying language model your agent will use. Cotonity currently offers three tiers: Standard (GPT-4o Mini equivalent — fast and cost-efficient), Advanced (GPT-4o equivalent — best for reasoning), and Expert (Claude Opus equivalent — highest accuracy). Each tier has different per-run costs shown inline. Next, write a system prompt in the text area provided. The system prompt defines your agent's persona, capabilities, and constraints. Be specific: tell the agent what it should and should not do, the tone it should adopt, and any output format requirements. You can use the built-in Prompt Improver button to have Cotonity's AI suggest refinements to your draft.
Add tools and knowledge sources
The Connect step lets you attach tools and knowledge sources. Tools are actions the agent can perform, such as sending an email, querying a database, or calling a webhook. Knowledge sources are documents or URLs the agent can retrieve information from at runtime — for example, a product FAQ PDF or a help-center URL. Click "Add Tool" to browse the built-in tool library or add a custom HTTP action. Click "Add Knowledge" to upload files (PDF, DOCX, TXT, CSV up to 50 MB each) or paste URLs to be crawled. You can also link an existing integration, such as a connected Slack workspace or a HubSpot account, as a tool target.
Test your agent before going live
The Test step provides a chat-style interface where you can send sample messages and inspect the agent's responses in real time. Each test run is free and does not count toward your monthly quota. Review the agent's replies for accuracy, tone, and format. If the agent makes mistakes, return to the Configure step and adjust the system prompt. Once you are satisfied, click "Publish Agent". The agent's status changes to Active and it becomes available to any workflow or integration that references it. You can always return to edit an active agent — changes take effect on the next run.