Local Language Models
The catalog of local Ollama text-generation models available for download from Settings → Agentic, grouped by what they're best at.
What "Local LLM" Means
Local Language Models run through Ollama, a separate server process installed and running on your own machine. Unlike In-Browser models, inference doesn't happen inside the browser tab itself — the browser talks to the Ollama server over HTTP on 127.0.0.1, and Ollama runs the model using your machine's full CPU/GPU resources.
Because Ollama is a standalone application, model downloads, storage, and GPU acceleration are all managed by Ollama itself — not by the browser tab or TotalApp's server. This means model sizes can be much larger than In-Browser models (several GB is common), and models persist on disk across restarts, independent of browser cache.
Where to configure it
Go to Settings → Agentic → Writer Engine, select Ollama, then pick a model from the built-in download list described below — or pull any model directly via the Ollama CLI and it will appear automatically. See Ollama Integration for how TotalApp connects to Ollama, and Agentic AI Settings for the full settings reference. For embedding models used by Semantic Search, see Embedded Models instead — text-generation and embedding models are configured separately. Don't have Ollama installed? See In-Browser Models for a no-install alternative that runs entirely inside the browser tab.
Underlying Technology
This engine talks to Ollama, a standalone local server application built on llama.cpp:
- Models are downloaded in GGUF format, Ollama's own model packaging built around llama.cpp's quantized weights.
- Runtime: llama.cpp, running as a local server process reachable over HTTP (
127.0.0.1:11434by default) — not inside the browser tab, so it can use your machine's full CPU/GPU without the memory limits of a JS engine.
See In-Browser Models (ONNX) or In-Browser Models (WebGPU) if you'd rather avoid installing a separate application.
General Purpose Models
The best balance of speed and quality for everyday chat, summarization, and general writing tasks:
| Model | Size | Best for |
|---|---|---|
mistral:latest | ~4.4 GB | An industry-standard, very stable and reliable general-purpose model. |
llama3.3:latest | ~5.1 GB | Meta's flagship model — the default most Ollama users reach for first, with strong multilingual (including Turkish) performance. |
gemma2:latest | ~9.6 GB | Google's next-generation model with strong general capability and language fluency. |
Reasoning & Heavy Tasks
For coding, complex math, logical reasoning, and deep analysis workloads:
| Model | Size | Best for |
|---|---|---|
deepseek-r1:7b | ~4.7 GB | A popular model with a "thinking" reasoning loop and strong problem-solving ability. |
qwen2.5:latest | ~23.9 GB | A high-parameter model built for heavy data processing and enterprise-grade tasks — needs a strong machine. |
phi4:latest | ~9.1 GB | Microsoft's model, optimized specifically for scientific, mathematical, and logical reasoning scenarios. |
Lightweight & Micro Models
For low-powered machines (no dedicated GPU) or very fast, simple background tasks:
| Model | Size | Best for |
|---|---|---|
llama3.2:1b | ~1.3 GB | Meta's ultra-lightweight model — great for basic text tasks at high speed on local devices without straining the machine. |
llama3.2:latest | ~2.0 GB | A slightly larger sibling of the 1B model, still light enough for everyday local use. |
phi3:latest | ~2.2 GB | A compact model that understands basic instructions without heavy RAM/CPU usage. |
Note on sizes
Sizes shown are approximate, rounded download sizes for each model's default Ollama build (typically Q4 quantization) — actual disk usage can vary slightly between Ollama versions.
Choosing a Model by Category
General Purpose
mistral, llama3.3, gemma2 — the safest default choice for everyday chat, summarization, and writing.
Reasoning & Heavy Tasks
deepseek-r1, qwen2.5, phi4 — best for code, math, and multi-step logical reasoning, at the cost of speed and disk space.
Lightweight & Micro
llama3.2:1b, llama3.2, phi3 — best for low-powered machines or fast, simple background tasks.
Downloading a Model
Alternatively, you can pull any model directly via the Ollama CLI on your machine (e.g. ollama pull llama3.3) — it will then appear automatically in the Settings dropdown alongside the built-in download list.
FAQ
mistral or llama3.3 are safe, reliable general-purpose defaults for most day-to-day writing and chat tasks.