Agent Configuration

There are several configurations for an agent which are available in the Agent Configuration tab in workflow editor.

Most of these are required to be configured only for advanced use-cases which require things like connectors for fetching data from external sources, custom nodes using third-party libraries, etc.

• Agent Interface

  The interface type of an agent defines how users interact with it. The Abacus.AI workflow editor supports various types of agents based on their interface:

  • DEFAULT: The standard interface for AI agents. Uses RJSF forms for taking inputs from users.
  • CHAT: Agents designed specifically for conversational interactions. Uses standard text box with option for attachments.
  • AUTONOMOUS: Agents that operate with minimal user intervention, making decisions and performing tasks independently. You can use trigger nodes to make a normal agent autonomous.

• Python packages

  Specify the Python packages and their versions required to execute the agent's code. No need to include standard python builtin modules here. Also we already support popular packages such as pandas and numpy, so there's no need to include them in package_requirements unless you require a specific version.

  Examples: pillow, nltk, wordcloud, reportlab, pyarrow, celery==5.1.2, nltk==3.6.2, geopandas==0.9.0

• Connectors

  Specify the connectors required for the agent to access external data sources for reading and writing operations. Users will need to authenticate these connectors before interacting with the agent. For example, to enable email access functionality, configure the Gmail connector with the appropriate authorization scopes.

  There are two types of connectors:

  1. User Level Connectors: Required for accessing data sources that are specific to the user interacting with the agent.
  2. Org Level Connectors: Required for accessing data sources that are shared across the organization.

• Modules

  Specify the custom modules to include in the agent's runtime environment. These modules must be previously registered in the platform and can be utilized within your workflow implementation.

• Common Source Code

  The code defined in this section can be used across all nodes in the workflow.

• Evaluation Feature Group

  Provides a way to validate the performance of your agent on a dataset that it has not seen during training, which is crucial for understanding how the agent might perform in a real-world scenario.