Class: LlamaBotRails::AgentStateBuilder

Inherits:
Object
  • Object
show all
Defined in:
lib/llama_bot_rails/agent_state_builder.rb

Overview

This state builder maps to a LangGraph agent state. Most agents will have custom state. You can create a custom agentstatebuilder when creating new, custom agents.

Instance Method Summary collapse

Constructor Details

#initialize(params:, context:) ⇒ AgentStateBuilder

Returns a new instance of AgentStateBuilder.



4
5
6
7
 
# File 'lib/llama_bot_rails/agent_state_builder.rb', line 4

def initialize(params:, context:)
  @params = params
  @context = context
end

Instance Method Details

#build ⇒ Object

Warning: Types must match exactly or you'll get Pydantic errors. It's brittle - If these don't match exactly what's in nodes.py LangGraph state pydantic types, (For example, having a null value/None type when it should be a string) it will the agent.. So if it doesn't map state types properly from the frontend, it will break. (must be exactly what's defined here). There won't be an exception thrown -- instead, you'll get an pydantic error message showing up in the BaseMessage content field. (In my case, it was a broken ToolMessage, but serializes from the inherited BaseMessage)



12
13
14
15
16
17
18
19
20
21
 
# File 'lib/llama_bot_rails/agent_state_builder.rb', line 12

def build
  {
    message: @params["message"], # Rails param from JS/chat UI. This is the user's message to the agent.
    thread_id: @params["thread_id"], # This is the thread id for the agent. It is used to track the conversation history.
    api_token: @context["api_token"], # This is an authenticated API token for the agent, so that it can authenticate with us. (It may need access to resources on our Rails app, such as the Rails Console.)
    agent_prompt: LlamaBotRails.agent_prompt_text, # System prompt instructions for the agent. Can be customized in app/llama_bot/prompts/agent_prompt.txt
    agent_name: "llamabot", #This routes to the appropriate LangGraph agent as defined in LlamaBot/langgraph.json, and enables us to access different agents on our LlamaBot server.
    available_routes: @context[:available_routes] # This is an array of routes that the agent can access. It is used to track the conversation history.
  }
end