Skip to content

Which AI models SiteMind uses, and where it uses none at all.

Most tools won’t tell you what’s running behind their answers. Here is the complete architecture: the two models involved, why they were chosen, and the specific instances where SiteMind serves instant cached responses without invoking an LLM.

The AI Models

Two models, two very different jobs

One converts your content and queries into dense 768-dim mathematical vectors. The other reads the retrieved grounding context and synthesizes the final reply.

Writing the answer

Google Gemini Flash-Lite

Google’s Gemini Flash-Lite is an ultra-fast, sub-second latency model engineered specifically for high-speed streaming and precision synthesis. Because our hybrid search has already retrieved the exact grounding context, Flash-Lite generates immediate, accurate answers with near-zero latency, ensuring visitor conversations feel instant and fluid with zero lag.

Understanding your content

gemini-embedding-001 (768 dimensions)

Turns every chunk of your content, and every visitor question, into a vector so the right passage can be found by meaning rather than keyword match.

Execution Path

What actually happens when someone asks a question

Five deterministic stages, in order. The language model appears only in step 4 and can be bypassed completely on exact semantic cache hits.

  1. 01

    Your visitor asks a question

    Deterministic / No model

    The question is converted into a vector using the embedding model, as a query rather than a document. The two are embedded differently, and mixing them degrades search quality.

  2. 02

    Search your content

    Deterministic / No model

    That vector is compared against every chunk of your indexed content to find the passages that genuinely match the meaning of the question.

  3. 03

    Nothing relevant found?

    Deterministic / No model

    The assistant returns a fixed "I couldn’t find that information on the website" and stops. No language model is invoked, so there is nothing available to invent an answer with.

  4. 04

    Build the prompt

    Deterministic / No model

    The retrieved passages are assembled into a prompt with instructions to answer only from that material, along with the last few turns of the conversation for context.

  5. 05

    Generate the answer

    Model call

    Only now does the chat model run. It reads the supplied passages and writes an answer from them, which is streamed back with citations linking to the pages the content came from.

Engineering Rationale

Why the model matters less than retrieval accuracy

“Which LLM does it use?” is the common question. What actually determines answer quality is hybrid chunk retrieval and strict grounding barriers.

Retrieval quality matters more than model choice

If the right passage reaches the model, a small model answers correctly. If it doesn’t, the largest model available will still get it wrong (it just sounds more convincing while doing so). That is why we invest in search quality rather than in a bigger model, and why "which LLM does it use" is a less useful question than it appears.

The model is swappable by design

Chat and embeddings both sit behind provider interfaces, so the underlying model can be changed without touching the knowledge pipeline. Each stored vector records the provider, model and dimensions that produced it, so a migration stays traceable rather than silently mixing incompatible embeddings.

A model call is not guaranteed

Most products send every question to the model no matter what. SiteMind treats the model call as the last step and skips it entirely when retrieval finds nothing: which is what makes an honest refusal structural rather than a matter of prompt wording.

Your content never trains anything

Your content is used to answer your own visitors and nothing else. It is not used to train foundation models and it is not sold. See the security page for the full data-handling detail.

Explicit Guardrails

What SiteMind does not let you do

Our honest technical boundaries, so you can evaluate fit with full transparency.

You cannot pick your own model today

SiteMind runs one configured chat model for everyone rather than offering a model selector. The architecture supports swapping it, but choosing per workspace is not something you can do from the dashboard. If you have a specific requirement here, talk to us.

It only knows what you have indexed

The assistant cannot answer from the model’s general world knowledge, by design. Anything not in your crawled pages, uploaded documents or connected sources will be declined rather than guessed at.

100% Managed AI infrastructure (Zero-BYOK required)

SiteMind is a hosted, fully managed product. We manage multi-key failover pools and absorb all model token and vector embedding costs directly. You do not need to provide external LLM API keys or maintain self-hosted model instances.

Model & Architecture FAQs

Test it on your own website in under 2 minutes.

Enter your domain to index your pages and preview live answers.

https://
No credit card required2-minute automated setupEmbed with one line