Agents

An agent is a prompt executor for tool execute_agent_query and others. Each agent can contain its own configuration, like model and temperature or even API endpoint, to send requests to user-deployed AIs. Currently, only OpenAI-compatible APIs are supported. The agent can be defined only for tools, and the default agent is used for executing the prompts from the quick action bar.

Data

The agent can be created or modified in Settings / Agents.

• Name is the name of the agent to refer to in tools like execute_agent_query

• API URL is the address of the agent API endpoint. Currently, only OpenAI-compatible APIs are supported. Use https://api.openai.com/v1 if not sure.

• API Key is the API secret token, and it can be found here. API key is stored in unencrypted form in the browser stored, and in encrypted form on the server.

• Prompt is the agent system prompt, which is the set of default instructions for an agent. The system prompt from settings is used when not set.

• Model to use with agent. gpt-4o is smarter, gpt-4o-mini is faster and cheaper. By default is gpt-4o.

• Temperature is the randomness of the responses of the model, which also can be thought as creativity. The advised value is 0 for the agents that are used for tasks to prevent randomness of interpreting the instructions.

• Max tokens is the maximum number of tokens in the response. This value limits the verbosiness of the agent and limits the amount of funds it burns for responses. OpenAI provides a pretty comprehensive guide on tokens.

• Headers is a list of headers to send to API. It needs to be used with custom-deployed agents.