Not an LLM. Not a chatbot. Not a framework.
There are three common confusions about IRIS. Worth clearing them. What follows compares IRIS against each approach in architectural, not commercial, terms.
| Dimension | Standalone LLM | Simple RAG | Agent framework | IRIS SCE |
|---|---|---|---|---|
| Factual verification | No | Limited | Variable | Own layer |
| Adversarial challenge | No | No | Manual | By architecture |
| Persistent memory | No | No | External | Core |
| Audit trail | No | Partial | Variable | Complete and mandatory |
| Explicit confidence score | No | No | No | Yes, every output |
| Internal data shield | No | No | No | Yes |
| Deployment sovereignty | No (external API) | Mixed | Variable | Full (private cloud, on-prem, edge) |
Why they are not the same
Standalone LLM vs IRIS
An isolated model answers with what it learned in training. It does not verify, remember or challenge. IRIS adds those three as architectural layers.
Simple RAG vs IRIS
RAG adds document retrieval, not deliberation. A retrieved claim is still an unchecked claim. IRIS verifies, challenges and consolidates.
Agent framework vs IRIS
An agent framework is an orchestration library. IRIS is an engine with proprietary cognitive architecture: memory, verification and audit trail belong to the engine, not to user code.
Enterprise chatbot vs IRIS
An enterprise chatbot is a conversational interface over a model. IRIS is the engine that could run behind; conversation is one channel among several.
The difference is the architecture, not the model.