Tool Integration

Julep provides a flexible system for integrating various types of tools that agents can use during interactions. These tools enable agents to perform actions, retrieve information, or interact with external systems.

Types of Tools

illustrates how to define a user-defined function and integrate it with Julep. Below is an example in YAML format:

name: fetch_user_profile
description: Retrieve user profile from the database.
parameters:
  type: object
  properties:
    user_id:
      type: string
      description: Unique identifier of the user.
    fields:
      type: array
      items:
        type: string
      description: Specific fields to retrieve from the profile.

You can use this function to call the relevant system tool for interacting with external databases or APIs.

1. User-defined Functions

   • Function signatures that you provide to the model

   • Similar to OpenAI's function-calling feature

   • Example:

     name: send_text_message
     type: function
     function:
       name: send_text_message
       description: Send a text message to a recipient.
       parameters:
         type: object
         properties:
           to:
             type: string
             description: Phone number of recipient.
           text:
             type: string
             description: Content of the message.
         required:
           - to
           - text

2. System Tools

   • Built-in tools for calling Julep APIs.

   • Can trigger task executions, append to metadata fields, etc.

   • Executed automatically when needed, no client-side action required.

   • Example:

     name: list_user_docs
     description: Get user docs
     type: system
     system:
       resource: user
       subresource: doc
       operation: list

3. Built-in Integrations

   • Integrated third-party tools from providers like wikipedia, brave and cloudinary.

   • Upcoming: Support planned for other langchain toolkits (Github, Gitlab, Gmail, Jira, MultiOn, Slack).

   • Executed directly on the Julep backend.

   • Additional runtime parameters can be set in agent/session/user metadata.

   • Example:

     name: internet_search
     description: Performs an internet search using Brave
     type: integration
     integration:
       provider: brave
       method: search
       setup:
         api_key: <BRAVE_API_KEY>

4. API Calls & Webhooks

   • Julep can build natural-language tools from OpenAPI specs.

   • Additional runtime parameters loaded from metadata fields.

   • Example:

     name: search_instagram_reels
     type: api_call
     api_call:
       method: GET
       url: https://instagram-scraper-api3.p.rapidapi.com/reels_by_keyword
       headers:
         x-rapidapi-key: <RAPID_API_KEY>
         x-rapidapi-host: <RAPID_API_HOST>
       follow_redirects: true

Partial Application of Arguments

Julep allows for partial application of arguments to tools using the x-tool-parameters field in metadata. This is useful for fixing certain parameters for a tool. Example:

{
  "metadata": {
    "x-tool-parameters": {
      "function:check_account_status": {
        "customer_id": 42
      }
    }
  }
}

Resolving Parameters with the Same Name

When multiple scopes (user, agent, session) define the same parameter, Julep follows a precedence order:

1. Session

2. User

3. Agent

This allows for flexible configuration of tools across different scopes while maintaining clear rules for parameter resolution.

By providing these various tool integration options and configuration capabilities, Julep enables the creation of powerful and flexible agent-based applications that can interact with a wide range of external systems and data sources.