Getting started
Contact your Palantir representative to install this beta product in your enrollment.
This guide demonstrates how to access AIP Agent Studio, introduces the AIP Agent Studio interface, describes how to set up a basic AIP Agent equipped with the information and tools that you choose, and how to deploy and monitor the AIP Agent in production.
Access AIP Agent Studio
AIP Agent Studio can be accessed from the platform’s workspace navigation bar or by using the quick search shortcuts CMD + J (macOS) or CTRL + J (Windows). Alternatively, you can create a new AIP Agent from your Files by selecting + New and then selecting AIP Agent, as shown below.
After opening AIP Agent Studio, you can create a new AIP Agent file.
Create an AIP Agent
AIP Agents are Palantir filesystem resources that have granular access control and can be created like any other filesystem resource, as shown in the image above, in the previous section.
You may also select the New AIP Agent option from within AIP Agent Studio.
Alternatively, create an AIP Agent from within AIP Threads.
Set up an AIP Agent
The following describes the setup for a "Standard agent", as opposed to an "AIP Assist Agent".
Add the name, description, and a photo as an avatar for your AIP Agent. This enables you to white-label your agent to fit the context of your application. If an avatar is not provided, a gray robot icon will be used as the default.
Depending on whether you finish the setup in the wizard or skip to the configuration panel, your creation workflow will be different. Either way, you will need to configure the enterprise-specific information and tools that will be equipped to your AIP Agent, as detailed in the following sections.
Types of information and tools
- Retrieval context: Simple and fast, recommended for most use cases.
- Application state: To contextualize agents in Workshop.
- Tools: Used in complex and action-taking agents. Slower time to first token.
These configurations are what enable the LLM to be useful to your enterprise, your workflow, and your task.
Choose a large language model (LLM)
The models available to you are a subset of those enabled on your stack.
Modify the system prompt
The system prompt should outline the AIP Agent's function within the context of the current application. By pressing / on your keyboard, you can refer to the configured tools and application state and guide the AIP Agent on how to coordinate their usage. Make sure to describe the underlying business logic and the appropriate situations for using the right tools in context.
Set the temperature
Users can modify the model temperature to determine the balance between focused, deterministic output (default value 0) and random output (maximum value 1).
Add conversation starters
You can also set up an input placeholder and suggested prompts to customize the agent for your intended workflow.
Save, view, and publish an AIP Agent
Once you have configured your AIP Agent, you can save your progress by using Save at the top right of the interface. To view your AIP Agent in action in the perspective of an end-user interaction, use View and select the desired version.
When you are ready to deploy your AIP Agent, select ✅ Publish to make your agent available for use in production environments. You can monitor the performance and usage of your agent through the Monitoring and Usage tabs, where you can see metrics and feedback from users.
Use in AIP Threads, Workshop, View mode, or OSDK with Developer Console and platform APIs.
Track AIP Agent feedback and usage
See how your AIP Agent is used in the Monitoring and Usage tabs of AIP Agent Studio, noting the feedback data. Feedback data is received from users giving thumbs-up or -down signs to the agent from a conversation.
Deploy an AIP Agent to a Developer Console application [Beta]
This feature is in beta and may not currently be available on all enrollments. Some functionality may change before this feature becomes generally available.
Once you have configured and published your AIP Agent, you can create and configure a Developer Console application to interact with your AIP Agent in a custom application.
To enable your Developer Console application to interact with your AIP Agent using platform APIs, follow the steps in create a new Developer Console application to create a new SDK application with access to Platform SDK resources:
To use an AIP Agent from an Ontology SDK application, you must configure the AIP Agent to only use object types, action types or function types from a single Ontology.
-
On the Resources page, select the Ontology used by your AIP Agent, then select all object types, action types and function types used in your AIP Agent configuration. Ensure you select the types used for the application state and all tools and retrieval context configured for your AIP Agent.
-
Next, select the Platform SDK tab. Under Projects access, add the project containing your AIP Agent. To find the project for the AIP Agent, open the AIP Agent in Agent Studio and inspect the filesystem path details next to the AIP Agent's name in the header.
-
If your AIP Agent is configured to use any other filesystem resources, such as a media set for document context, ensure these are in the same project as your AIP Agent, or add all additional projects for these resources to the Projects access section.
- Finally, enable operations for the AIP Agents API in the Client allowed operations table. To check the operations required for the different AIP Agent platform APIs, refer to the platform API documentation.
To allow your Developer Console application to create and send messages in conversation sessions with an AIP Agent, ensure you enable the AIP Agents write permission.
- Refer to the Developer Console documentation steps to review and finalize creation of your application.
Update an AIP Agent used in Developer Console applications
Once you have configured a Developer Console application to allow interaction with an AIP Agent through Platform SDK resources, you will need to update the application if any of the Ontology or platform resources used by your AIP Agent are modified.
For example, if you add any new object types, action types or function types to the AIP Agent, you must add these to the Ontology SDK resources for your application in Developer Console. Similarly, if you add any platform resources to your AIP Agent, such as additional media sets for document context retrieval, you must add these to the Platform SDK resources for your application. Developer Console application resources are not updated automatically when changes occur to the types of resources used by your AIP Agent.
Create conversations with AIP Agents in custom applications
To get started bootstrapping a new application, refer to the documentation examples for TypeScript or Python, or add the SDK to an existing application.
Once you have created your application, use the Create Session platform API to create a new conversation with your AIP Agent.
The Sessions APIs for AIP Agents require you to specify the agentRid for the AIP Agent to use for conversation session interactions.
You can find this by opening the project for your AIP Agent, selecting the AIP Agent and using the Copy to clipboard option for the RID under Metadata in the file overview.
Once you have created a new session, use the sessionRid value in the returned response to send a new message to the AIP Agent and get responses using the Blocking Continue Session or Streaming Continue Session platform APIs. Use the blocking API to wait to receive the full AIP Agent response once it is fully generated, or use the streaming API to receive a stream of the AIP Agent's answer text as it is generated.
You can load conversation metadata for a session using the Get Session API, and load the history of exchanges (messages sent by your application and responses from the AIP Agent) for a session with the Get Content API.
Refer to the platform API documentation for code examples on how to use these APIs in your target application language.