​
Leveraging an agent

Build an agent Prompts are the main way to give immediate instructions to agents. These can either be user-generated or standardized inside a play. A prompt can be passed to a play via the agent’s chat editor or via a trigger from one of its deployments. The system prompt defines the agent’s role, capabilities, and constraints

​
Agent components

Agents need access to tools and resources to function and are powered by foundational LLM models that power their semantic reasoning. Tools Equip an agent with a tool Tools equip agents with specific capabilities, such as API access to external systems and their responses. These act as the agent’s “hands” for interacting with external data sources and services. Resources Add a resource to a model Resources are the data sources and knowledge bases that agents can draw from. This can be any data model in your workspace, such as your CRM, a warehouse table, or external databases. AI models Add a model Agents are powered by AI models that provide the reasoning and decision-making capabilities. You can choose between OpenAI or Anthropic models depending on your specific needs and requirements. Configuration parameters Agents use two main parameters to control their behavior.
  • Reasoning steps define the number of logical steps the agent can take to conclude.
  • Temperature: controls the creativity and randomness of responses.

​
Agent deployment

Once created, agents can be deployed across multiple environments and platforms to integrate with your existing workflows and systems. Slack integration Deploy a model to slack Agents can be deployed directly in Slack, allowing team members to trigger them via the ‘@Cargo’ tag. Play integration Agents can be embedded within plays as intelligent nodes, providing non-deterministic research and decision-making capabilities to deterministic workflows. Browser extension Agents can be triggered via Cargo’s chrome extension, enabling the agent to access real-time information from the user’s browser. API deployment Agents can be triggered via REST API endpoints, allowing external systems and applications to access them.