Create Custom Agents
Custom Agents are specialized AI assistants that you can customize to meet your exact needs. Visit you.com/agents to create one. Then, you’ll be able to use our Custom Agents API to interact with it.
Configuring your Custom Agent
Name & description
Give your agent a clear name and description that explains its purpose. These are for usability only and do not affect behavior.
Instructions
Instructions are the core of your agent—they define its behavior, personality, output format and more. Examples of what you can control include:
- What the agent does - Its role, purpose, and scope
- How it responds - Tone, style, language, verbosity and level of detail
- Output formatting - How to compose and structure its response
- Guidelines and constraints - What to prioritize, avoid, or handle specially
Don’t be afraid to be detailed. The more specific your instructions, the more effective and repeatable your agent’s behavior will be.
Response model
Choose the model that powers your agent. This also weighs signifcantly on your agent’s performance.
- Auto - Automatically selects the best underlying LLM based on the query
- Express - Optimized for speed
- ARI - Advanced Research & Insights. This is our highly intelligent deep research agent. Learn more
- Foundational models - Choose from the best foundational LLMs from OpenAI, Anthropic, Google and more
Sources
Customize what data and resources your agent grounds its answers in.
Files & links
Attach files or specify web URLs for your agent to ground its responses in. You can upload your own files, or attach directly from Google Drive and OneDrive.
Currently, files that you attach are not visible to Custom Agents used via API
Web search
Your agent can pull real-time information from the internet if you so chose. You can adjust the search effort level to balance between speed and depth. You can also exclude specific web sources.
Share access
Control who can see and access your agent. Remember that attached sources, files, and instructions can be queried by anyone with access.
Advanced research & reasoning
Your agent can research multiple sources and reason through complex queries in multiple steps. This is only available with certain response models.
- Auto: Can be toggled on/off
- Express: Not available. Express is designed for speed
- ARI: Always enabled and cannot be disabled. It is core to ARI’s behavior
- Foundational model: If your agent uses a foundational model (for example, Anthropic’s Claude Sonnet 4.5), advanced research & reasoning is not supported
Using your Agent via API
Once created, your agent will appear in the Agents section on the platform. Grab its ID and check out the Custom Agent API to learn how to interact with it.
