Agents

An agent is essentially an AI assistant with a knowledge base and a specific set of capabilities. Technically, an agent is powered by a set of models (typically large language models) from a particular AI platform (OpenAI, Google AI, etc.) working together towards a goal, using the custom data and functions available to them. For instance, an OpenAI-powered agent may use gpt-4o for chat completion and whisper-1 for audio transcriptions.

Currently, most agents use OpenAI assistants behind the scenes. Whenever you create an agent on ProteusAI, a corresponding assistant is created in your OpenAI Playground (opens in a new tab). You can also attach a new ProteusAI agent to an existing OpenAI assistant.

More AI platforms will be supported in the future.

Click the New Agent button and fill in the required fields in the pop up modal.

This opens the agent configuration page with 4 important menus: Instruction, Knowledge, Customizations, and Integrations.

Test Agents

It is now possible to create a free test agent which does not require you to provide API credentials. To do so, simply turn on the Create as free test agent option. Each account is entitled to a maximum of one test agent.

Computationally-Optimized Agents

It is now possible to create an agent that is optimized for computation to enable features like advanced math calculations and code execution. To do so, simply turn on the Optimize for computation option. This behavior cannot be changed after the agent is created.

Computationally-optimized agents are able to iteratively solve challenging code and math problems. They typically do so by writing and running Python code in a sandboxed execution environment, and can generate files with data and images of graphs.

Computationally-optimized agents can accept these file types (which other agents may not be able to handle):

• .csv
• .jpeg, .jpg
• .gif
• .png
• .pkl
• .tar
• .xlsx
• .xml
• .zip