Pāriet uz galveno saturu

Build and publish Atlas AI agents

Early adopters

The features described in this section are currently available to early adopters only and are subject to change.

Build and publish Atlas AI agents to solve business problems or automate workflows using the low-code Agent builder. You can build your agent from scratch, or select an out-of-the-box template as a starting point.

Iterate rapidly with the Agent builder chat preview. When you're satisfied with the agent's performance, publish it and make it available in the Agent library to all users in the CDF project.

Before you start

Before you start building the agent, scope your use case, identify an evaluation data set, and choose the language model to start with.

Build an publish an agent

To build and publish an Atlas AI agent:

  1. Navigate to Atlas AI > Agent builder.

  2. Select + Create agent or select a template to use as a starting point.

  3. Enter a Name for the agent.

  4. Enter a Description to help users understand what they can use the agent for, for example, which problems the agent can help solve or which tasks it can automate.

  5. Enter one or more Sample prompts to display as examples to users when they interact with the agent.

  6. In the Prompting section:

    1. Select the Language model you have identified as the best fit for your agent.

    2. Specify Goals to define the desired outcomes or objectives of the interactions with the agent. Goals outline what you want to accomplish.

    3. Provide Instructions with specific directives or guidelines for how the agent should achieve the goals.

      See Prompts and prompt engineering for more details.

  7. Add Tools to allow the agent to access data in the CDF knowledge graph, perform more complex tasks, or interact with other applications.

  8. Use the chat interface to test the agent. To refine the agent, adjust the language model, goals, instructions, and tools as necessary.

  9. Select Publish to make the agent available in the Agent library to all users in the CDF project.

  10. Monitor the performance and effectiveness of the agent and make ongoing improvements.