Agents without LLM usage

[kaikreuzer]

I would like to implement a kind of "proxy agent", i.e. one that simply forwards the incoming question to an external chatbot and then returns the received answer back as a response.
I assume Arc is not the right framework to implement this, so it's rather a "bring your own agent" (BYOA) type of problem. Are there already any best practices on how I could build such an agent for LMOS?

[RobWin]

The current LMOS Runtime communicates with Arc agents using GraphQL WebSockets, but there are plans to transition to a Web of Things (WoT) communication protocol. This transition will offer greater flexibility by enabling protocol bindings like HTTP, MQTT, AMQP, or WebSockets. We have two options:

Option 1: Implementing the Proxy with the WoT Approach

1. Use the WoT Library:
   The upcoming Web of Things Kotlin library will allow you to implement a proxy agent that provides a WoT-compliant interface for the external chatbot.

2. WoT Description:
   The proxy will expose a WoT description, detailing the proxied agent’s capabilities. The LMOS runtime can interact with this proxy using a WoT client.

3. Protocol Bindings:

   • The library supports multiple bindings (e.g., HTTP, MQTT).
   • While the HTTP binding is currently functional, work is ongoing for the MQTT binding. You could start a proof of concept (PoC) using HTTP and later migrate to MQTT or another protocol as needed.

4. Advantages of WoT:

   • Standardized and interoperable communication.
   • Flexibility in switching protocols based on requirements.

---

Option 2: Using the Current GraphQL Approach

1. Spring Boot Application:
   Develop a Spring Boot application that acts as a GraphQL-to-HTTP proxy. This application would:

   • Receive GraphQL queries from the LMOS runtime.
   • Translate these queries into HTTP requests for the external chatbot.
   • Process the responses from the chatbot and return them in GraphQL format.

2. GraphQL Spring Boot Starter:
   Use the GraphQL Spring Boot Starter to simplify the implementation of the GraphQL API for the LMOS runtime.

3. Advantages of GraphQL Approach:

   • Direct compatibility with the current LMOS setup.
   • No need to modify the runtime immediately.

---

Recommendation

• If you're looking for a quick and temporary solution, go with Option 2 (GraphQL proxy).
• If you prefer a future-proof and standards-compliant solution, start a PoC with the WoT library using the HTTP binding, and adapt the LMOS runtime to use a WoT client.