Build a new agent

Get started quickly with our builder agent setup using this simple guide. The guide will walk you through the creation of a new agent that can be embedded into your webpage using iframe or plain REST API (using SUPERWISE® SDK).

Creating an agent can be done using the SUPERWISE® UI console or using REST APIs or SDK.

Using the UI:

Creating the agent requires a few simple steps:

• Create an agent - In the Agents page, press the “Create” button. Give it a name, select the agent type and you will be directed to the agent builder studio.
• The first step to configure your agent is to configure the underlying LLM model. Once you configure your model, the agent playground will be enabled, in which you can experiment with your agent interaction.
• If your in AI-Assistant-Retrieval agent type, you must connect a context to your agent.
• Once your agent is ready to go, you can “Publish” it, and then the agent will be ready to be embedded. To read more on how to embed your agent, please visit the Embed your agent guide.

To read more on how to configure and further fine-tune your agent, visit our Build an agent set of guides.

Using the SDK - Basic LLM agent example

📘

Prerequisites

To install and properly configure the SUPERWISE® SDK, please visit the SDK guide.

In this example we will create a Basic LLM agent. Read on more agent types here.

Create and deploy an agent

from superwise_api.models.agent.agent import BasicLLMConfig
from superwise_api.models.agent.agent import OpenAIModel, OpenAIModelVersion

# Create an agent
agent = sw.agent.create(f"my agent name", description="Description for my agent", authentication_enabled=False, observability_enabled=True)

# Create a model
model = OpenAIModel(version=OpenAIModelVersion.GPT_4, api_token="Add your API token")

# Assign agent configuration
agent_config = BasicLLMConfig(llm_model=model, prompt=prompt)

# Publish the configuration as a new version
version = sw.agent.create_version(agent_id=agent.id,name="My Version name", description="My Version description", agent_config=agent_config)

If you want to see what models are available you can read here for more.

Use your agent

Now that your agent is published, ask him a question:

agent_response = sw.agent.ask_worker(agent_id=agent.id, input="your question here")

print(agent_response)