Agentic & AI Architecture

How TotalApp turns your data into grounded AI answers — the models and runtimes that do the work, and the engine that finds, ranks and reasons over your records.

Overview

This section documents everything behind TotalApp’s AI features, in four parts. AI Models & Runtimes covers where the work physically happens — on your own machine, inside the browser tab, or on a cloud provider. Data & RAG covers how TotalApp finds the right records and puts the best one first. AI Assistants covers the panels and screens where you actually work with the AI. Security & Governance covers how the pieces are reasoned over, chained together, protected and kept under control.

The core idea

Facts come from your data; the model only reasons over what was found. Every step narrows the field, and every step has a fallback so a missing model never breaks a screen.

From question to answer

A typical AI feature moves through the stages below. Not every feature uses every stage — exact lookups skip the semantic ones, and reranking is optional.

Question

→

Exact lookup or Embedding

→

Reranking (optional)

→

AI reasoning

→

Governed answer

Stage	What it does	Where it runs	Learn more
Exact lookup	Finds records with traditional, 100% reliable database queries	Your tenant’s data	2-Tier Architecture
Embedding	Shortlists the records closest in meaning to your text	Local Ollama or a hosted model	Embedding
Reranking	Re-scores the shortlist so the best match comes first	A hosted rerank model	Reranking
AI reasoning	Interprets, audits or summarizes the records it was given	The AI engine you chose	AI Models & Runtimes
Orchestration & governance	Chains agents together and controls who can do what	TotalApp	AI Orchestration, AI Governance & Access, AI Security

Where you configure it

Everything is configured from Settings → Agentic, which is split into sub-tabs so each kind of model has its own place. The most relevant ones for this section:

Sub-tab	What it controls
Chat Model	The AI engine that generates text: Ollama, an in-browser model, or a cloud provider
Embedding	Where Semantic Search gets its vectors: local Ollama or a hosted model
Rerank	The optional model that re-scores the best Semantic Search matches
General	AI usage on or off, default model, limits, retries and the AI Governance policy

These choices are independent. You can, for example, keep embeddings on your own machine, generate text with a cloud model, and leave reranking off. See Agentic AI Settings for the full reference.

Local or hosted?

Each model can run on your machine or at a provider. The trade-off is control versus convenience.

	Local (Ollama, in-browser)	Hosted (Cloud LLM)
Where your text goes	Stays on your machine	Sent through the TotalApp server to the provider you chose
Setup	Install Ollama or load a browser model	Pick a model from the list — nothing to install
Cost	Free	Free models available; paid models billed by the provider
Works offline	Yes	No

Reranking is hosted-only. If your text must not leave your environment, use local models and leave Reranking unselected.

Built to degrade gracefully

AI should never be the reason a screen stops working. If a layer is unavailable, TotalApp falls back to the one below it:

No rerank model selected, or the service is unavailable: the normal embedding order is used
No embedding model, or it cannot be reached: simple keyword matching still ranks your records
No AI model: tables, forms and every non-AI feature keep working — only the AI buttons are affected

Pages in this section

AI Models & Runtimes

Overview

The five engines and how they fit together.

Cloud LLM Models

Hosted providers and their models.

Local Language Models

Downloadable models for Ollama.

Ollama Integration

How the browser talks to your local Ollama.

In-Browser Models (ONNX) · (WebGPU)

Models that run inside the browser tab.

Embedded Models

The catalog of embedding models.

Data & RAG

Embedding

Finding records by meaning.

Reranking

Putting the best match first.

AI Assistants

Overview

Where the assistants live across the app.

AI Assistant Panel

The side panel for asking and acting.

Table & List Assistant

Ask questions about the rows you see.

Dashboard Assistant

Summaries and answers on dashboards.

Security & Governance

AI Governance & Access

Access, topic rules, PII masking and audit.

AI Security

Prompt protection, file handling and isolation.

AI Orchestration

Building and running agent pipelines.

2-Tier Architecture

Deterministic data first, AI reasoning second.

Frequently Asked Questions

Do I need to set up every layer?

No. Exact lookups and every non-AI feature work with nothing configured. Add an embedding model when you want Semantic Search, a chat model when you want AI answers, and a rerank model only if you want sharper ranking.

Does Reranking need Embedding?

Yes. Reranking re-scores the shortlist that Embedding produces, so an embedding model must be selected first.

Can I mix local and cloud models?

Yes. The AI engine, the embedding model and the rerank model are chosen independently.

Where do governance and orchestration fit?

They sit around the pipeline: orchestration chains agents together, and governance controls access, topic and personal-data rules and the audit trail, while AI Security describes how requests and files are protected on the way.