Agentic Overview
How TotalApp routes agentic text generation and embeddings across five engines — Ollama, local Ollama models, in-browser models, cloud APIs, and embedding models — all configured from Settings → Agentic.
What Lives Under Agentic
The Agentic section documents every engine and model catalog available from Settings → Agentic. Two independent choices exist on that screen: the AI Engine (what generates text) and the Embedding Engine (what powers Semantic Search ranking) — they're configured separately and can be mixed freely.
Where to configure it
Go to Settings → Agentic to choose an AI Engine and, independently, an Embedding Engine. See Agentic AI Settings for the full settings reference.
The Four AI Engines
Choosing an AI Engine decides where your text-generation requests actually run:
| Engine | Runs on | Requires | Docs page |
|---|---|---|---|
| Ollama | Your own machine (local server process) | Ollama installed & running | Ollama Integration, Local Language Models |
| In-Browser LLM (WebGPU) | The browser tab itself (WebGPU/WASM) | Nothing — model downloads once and is cached by the browser | In-Browser Models |
| Cloud LLM (Hosted API) | A third-party provider's servers | An API key per provider | Cloud LLM Models |
| Local CLI | Your own machine, via the installed claude CLI | Claude CLI installed & authenticated | Agentic AI Settings |
Ollama Integration
TotalApp integrates with Ollama, a local model runner — both text generation and embeddings can run entirely on your own machine, with no traffic ever reaching TotalApp's server. See Ollama Integration for the full explanation of how the browser talks directly to your local Ollama instance.
Local Language Models (Ollama)
Once Ollama is running, Local Language Models catalogs the downloadable text-generation models, grouped by what they're best at:
General Purpose
mistral, llama3.3, gemma2 — the safest default for everyday chat and writing.
Reasoning & Heavy Tasks
deepseek-r1, qwen2.5, phi4 — code, math, and multi-step logic.
Lightweight & Micro
llama3.2:1b, llama3.2, phi3 — low-powered machines or fast background tasks.
In-Browser Models
No Ollama install needed at all: In-Browser Models run directly inside the browser tab's own memory and GPU (via WebGPU, falling back to WASM), using Transformers.js. Nothing is installed outside the browser.
Micro / Low-Resource
SmolLM2-135M, SmolLM2-360M, Qwen2.5-0.5B — old or GPU-less machines.
Balanced In-Browser
TinyLlama-1.1B, Llama-3.2-1B-Instruct, SmolLM2-1.7B — everyday in-browser writing and chat.
High Performance
Qwen2.5-1.5B-Instruct, Qwen2.5-Coder-1.5B — best in-browser reasoning and code quality.
Cloud LLM Models
Cloud LLM Models catalogs every hosted provider, grouped by architecture:
| Architecture | Providers | Notes |
|---|---|---|
| Direct | Anthropic, OpenAI, Google AI Studio, DeepSeek | Lowest latency/cost for that provider's own models. |
| Proxy / Router & Serverless | Cohere, Groq | Managed infrastructure; Groq specializes in very high inference speed for open models. |
| Aggregator | OpenRouter | One API key, many providers — usually a small markup over calling directly. |
Embedding Models
Independent of the AI Engine, TotalApp uses local Ollama models exclusively for embeddings — used to rank and retrieve knowledge sources for Semantic Search. See Embedded Models for the full catalog, recommended starting points (nomic-embed-text for long English context, bge-m3 for multilingual/Turkish), and download instructions.
Choosing an Engine
Want full privacy, have a capable machine?
Use Ollama with a model from Local Language Models.
Want zero install, quick trial?
Use In-Browser Models — nothing to set up outside the browser.
Want the strongest models available?
Use Cloud LLM Models with an API key from your preferred provider.
FAQ
claude CLI — no separate server to manage.