Learn Facts from Users

Learn facts from your users and save them to a vector database

You can use the VectorSearchLearnerTool (opens in a new tab) to learn facts from users and save them to a vector database.

Using this tool only requires providing it to your FunctionsBasedAgent (opens in a new tab):

self._agent = FunctionsBasedAgent(
    tools=[
        VectorSearchLearnerTool()
    ],
    llm=OpenAI(self.client),
)

By default, the tool will:

• Use your agent's default vector database, created on demand for you inside Steamship
• Ask the agent to save facts whenever the user requests that the agent learn or remember something

To customize these to behaviors, just provide these arguments to the tool's constructor:

• index_handle - The name of the vector database to use (created on-demand for you; must be a-z and dashes)
• ai_description - Tells your agent how and when to use the tool
• name - The name of this tool; only important if you plan to create multiple copies of the tool

For example, to add a VectorSearchLearnerTool that records your likes and dislikes to a specific vector database, you might say:

VectorSearchLearnerTool(
    name="PreferenceLearner",
    index_handle="personal-preference-index",
    agent_description = (
        "Used to remember a personal preference. Use this tool every time the user expresses a like, dislike, or preference. ",
        "Use this tool even if the user did not mention they want something remembered. ",
        "The input is a personal preference to learn. ",
        "The output is a confirmation of the personal preference that will be remembered.",
    )

Building an Agent with VectorSearchLearnerTool

This tool can be provided to a FunctionsBasedAgent (opens in a new tab) which will decide when to use it based on its ai_description. For example, if the tool's ai_description says the tool is "Used to remember a fact. Only use this tool if someone asks to remember or learn something", then the agent will tend to use the tool when the user makes a statement like "Remember that Susan lives in Boston."

In your AgentService, here's how you can provide this tool to a FunctionsBasedAgent (opens in a new tab):

# Create the AgentService's agent, providing it the tool
self._agent = FunctionsBasedAgent(
    tools=[
        VectorSearchLearnerTool(),
    ],
    llm=OpenAI(self.client),
)

A full, working example can be found at TODO.

Deploying an AgentService with VectorSearchLearnerTool

To deploy an agent with this tool, just make sure it's properly wrapped in an AgentService wrapper and then follow the normal deployment instructions. You will get a URL that you can visit to create instances of your agent in the cloud and connect to it from the web & chat apps.

Customizing the Behavior of VectorSearchLearnerTool

Steamship tools are just open-source Python. The source for this tool is available here (opens in a new tab). You can customize it by copy-pasting the tool into your own project, changing the code, and then using your new version.

This particular tool works by:

1. Loading an Embedding Index Plugin from Steamship (this auto-manages embeddings & vector databases for you)
2. Adding the provided text to the Embedding Index
3. Responding with templated statement that the new facts will be remembered

A few ways you might customize it might be:

• Apply a prompt before saving it to the embedding index, using the LLM in the provided AgentContext. Perhaps the prompt can rewrite, summarize, or expand upon the provided information.
• Add additional information to the metadata that is saved with the embedded item. Perhaps context in which the fact is being learned.

If you create an interesting customization of this tool, please consider sending us a pull request on GitHub (opens in a new tab)!