Vault AI Agents

Platform Vault AI allows Admins to configure custom agents with direct, secure access to Vault object and document data. Agents use a Large Language Model (LLM) to allow them to perform a set of agent actions. Depending on your Vault application and configuration, users can execute agent actions through Vault AI Chat or directly through the Vault UI.

With Vault developer tools, developers can:

Execute agent actions with Vault API or Vault Java SDK
Create custom context types
Create custom agent tool types

Glossary of Vault AI Terms

Before developing with Vault AI, you should be familiar with the following terms and concepts:

Agent: Contains the agent configuration definition, which includes the set of available agent actions, the agent objective, agent context, and any configured agent tools. You can configure custom agents in the Vault UI. Agents are Vault-specific, meaning they cannot interact with data on other Vault domains or outside of your Vault. Each agent has a scope that determines the context it requires to execute:
- Document agents: Operate on a specific document and require a document ID to execute.
- Object agents: Operate on a specific Vault object record and require an object name, object type, and record ID to execute.
- Global agents: Are not tied to a specific document or object and do not require a document ID or object record ID to execute. Global agents can only be created by Veeva and are not available for customer configuration.
- Agent Objective: Defines the agent’s roles and responsibilities, similar to a job description.
- Agent Context: Defines the information an agent knows about your Vault, allowing the agent action to use this information when creating a response. Agent context can include both object and document metadata and document content. Context types are scoped to match the agent type: document, object, or Global. With Vault Java SDK, your organization can create custom context types.
Agent Action: The actions users can execute with Vault AI, such as Summarize Document and Translate Record. Agents perform agent actions based on the provided agent objective, context, and instructions. If the action is configured with supportChat = true, users can execute the agent action in the Vault UI through Vault AI Chat. Developers can execute agent actions through Vault API or Vault Java SDK.
- Agent Action Description: Defines the purpose of the agent action. While the primary goal of the description is to provide the purpose of this action to users and Vault Admins, Vault sends this information to your LLM to help it understand the agent action.
- Agent Action Instructions: Defines how to run the agent action. Your LLM uses these instructions to understand how to prepare and present output for the action.
Agent Tools: Agent tools provide your agent with additional information to execute an agent action. When executing an agent action, your LLM can choose to use a configured agent tool and use the tool's response to complete the action. You can configure custom agent tools in the Vault UI.
- Tool Types: Agent Tool Types define a tool’s function, which defines the information a tool can provide. For example, the VQL Query tool type allows an agent action to execute preconfigured VQL queries. With Vault Java SDK, your organization can create custom agent Tool Types.