Does anyone have a detailed example or tutorial on how to create an AIP Agent? I’m looking to learn the process and best practices for setting up an agent using Agent Studio. A step-by-step walkthrough or use case example would be highly appreciated!
1 Like
We have docs here yet https://www.palantir.com/docs/foundry/agent-studio/overview
We don’t currently have any examples to install via Examples Build with AIP. We plan to get those out soon though since AIP Agents now have Marketplace support, which backs those examples.
Are there specific questions you have?
2 Likes
Hi @AryanThadani02 !
Like @narmbrust mentioned, I would strongly recommend reading through the docs to get a better understanding of the different use cases for AIP Agents.
To give you a high level breakdown…
- Retrieval Context: This can be added to your agent if you would like to interact with enterprise-specific data, and not just the data that is native to the LLM. For example, you could upload PDF documents (such as manuals, guides, articles, etc.) so that your agent can answer questions with citations from the document material. If you have valuable data within your ontology (such as product feedback, support issues, etc.), you can also configure your agent to use ontology objects as context. For more advanced use cases, you can enable your agent to use function-backed context (coming soon to the documentation), where you write your own retrieval function in Python, Typescript, or AIP Logic that returns a string of context for the agent to reference.
- Application Variables: These can be added onto your agent so that it can read from and write to a global application state. This is especially useful for Agents that are consumed in Workshop via the AIP Interactive Widget. For instance, if you want to create an application where a user can filter a set of support tickets and use an agent to help resolve each one, you would simply add the object set of support tickets as an application variable to your agent. This enables the agent to access and interact with the tickets effectively.
- Tools: These are functions and actions that the agent can invoke whenever it determines they are necessary to answer a question. For example, if you want an agent to read a support ticket and mark it as resolved, you would configure an action on the agent that edits the support ticket ontology object.
Hopefully that gives you enough to get started, but don’t hesitate to reach out if you have any questions specific to your use case!
2 Likes