Local Contextual RAG for Theia AI

[fayakun-it-consulting]

With the actual state of affairs concerning Theia AI, is there any possibility to use the following functionalities?

1. Contextual RAG for complex document understanding
2. Memory context for long term context retention
3. Long term memory retention between AI Agent collaboration?

Is there any mechanism available for localhost document embedding/retrieval let's say when I implement and Agent who is specialized in researching with PDF documents & online web site?

[planger]

While there aren't any off-the-shelf components in Theia AI directly that provide those functionalities out of the box, you can indeed implement / integrate them in Theia AI. Good starting points for that are:

• Integrating RAG via variables: By defining your custom variable, you can integrate a RAG process, e.g. by invoking an external RAG component and returning its most relevant matches for the user context as a variable result, which then will replace the occurance of the #variable position in the prompt (whether it is a user message or system message).
• Tool calls or MCP tools enable the LLM to query or write to memory databases autonomously. There are off-the-shelf MCP servers for memory management. You can integrate those using the MCP capabilities of Theia AI.
• Creating your custom chat agent allows you to entirely take over the request from the user and augment it with whatever context you may need programmatically before you hand it off to the LLM. This way you can integrate your RAG component, query memory, perform arbitrary prompt flows, or even delegate to an agent that runs externally.

[fayakun-it-consulting]

Theia AI actual context window is only what you are providing it during the 1:1 chat prompt am I correct?

[planger]

The context is entirely controlled by the agent (getMessages and system message). The default implementation adds the configured system message of the respective agent and the conversation history so far.

[fayakun-it-consulting]

There's no such function as @AgentSmith can you regurgitate this web site documentation and help me find a solution for problem X?