Beginner's Guide

This guide walks you through everything you need to start using InControl, from installation to your first conversation.

What is InControl?

InControl is an Ollama chat application for Windows. The Microsoft Store lists it as InControl-Desktop. Chat stays on this PC until a rented GPU is connected. Nothing is sent to a rental until you connect one. A rental sends the prompts you submit to Ollama on that machine, and the banner names that machine while those prompts leave.

Prerequisites

Before installing InControl, make sure you have:

Windows 10 version 2004 or later, x64 (Windows 11 recommended)
A GPU Ollama can use — for example an RTX 3060 (8 GB VRAM) at minimum, an RTX 4080/5080 (16 GB VRAM) recommended. Without one, Ollama runs on the CPU, slowly. A rented GPU is the other option.
Ollama — the local LLM backend. Install from ollama.com/download

Installation

Microsoft Store (recommended)

Open InControl-Desktop in the Microsoft Store and choose Get. The package includes everything it needs, including .NET, and the Store keeps it updated.

Build from source

You need the .NET 9 SDK. Clone the repository, run the tests, and start the app:

git clone https://github.com/mcp-tool-shop-org/incontrol.git
cd incontrol
dotnet test tests/InControl.Core.Tests
dotnet run --project src/InControl.App

First-run walkthrough

When you launch InControl for the first time, the onboarding wizard guides you through four steps:

Welcome — introduces InControl. Chat stays on this PC until a rented GPU is connected.
Backend Check — InControl looks for a running Ollama on 127.0.0.1:11434. If Ollama is not running, start it with ollama serve in a separate terminal.
Model Selection — pick a model from those available on your Ollama instance. If you have no models yet, pull one first: ollama pull llama3.2.
Ready — you are all set. Click “Start” to begin your first conversation.

You can skip onboarding if you prefer to configure things manually.

Core concepts

Conversations and sessions

Each conversation is stored as a session file in %LOCALAPPDATA%\InControl\sessions\. Nothing is sent to a rental until you connect one. A rental sends the prompts you submit to Ollama on that machine. The banner names that machine while those prompts leave. Right-click a session to rename, duplicate, pin, export it as JSON, or delete it.

Projects and notes

Sessions live in projects. General is where a session goes by default. A project can carry instructions that are added to every message in it. Right-click a session to remember a note for that session or its project. Notes that match a new message are sent along with it, so they reach a rental too while one is connected. Clear All Memory in Settings deletes every note and keeps the chats.

Voice

InControl can read replies aloud with Kokoro, which runs on this PC. The first time voice is needed, when a reply is spoken or you open Settings, it downloads the voice model, about 300 MB. Turn off auto-speak in Settings if you don’t want replies read aloud.

Web search

The Web button next to the paperclip lets the model search the web. It is off until you turn it on, and the choice is remembered. While it is on, a model that can use tools (gemma3, qwen3, llama3.1 and others) can search DuckDuckGo and read public pages, and each reply lists what it searched. A model without tool support answers without searching, and the reply says so. Offline mode turns web search off. Search queries leave this PC, so leave Web off for anything private.

Attaching files

Click the paperclip next to the model picker, press Ctrl+Shift+O, or drop files on the composer. Each file shows as a chip; click it to remove it before sending.

Text and code files (up to 256 KB each) go into the message in a fenced block, so any model can read them.
Images (PNG, JPEG or WebP, up to 10 MB) go to the model beside the message. Only a vision model, such as gemma3 or llama3.2-vision, can read them. If the selected model cannot, InControl says so before sending and keeps the image attached.

Attached files are saved with the chat. While a rental is connected, they go to that machine with the message.

The offline switch

Offline is off when the app is installed. Turning it on refuses a rented GPU, a RunPod lookup, and a model download. Chat on this PC still works. The switch does not stop the voice model download. It is not a kill switch for every socket, and it is not the default.

Tool-URL allowlisting is a different control. See the repository file docs/CONNECTIVITY.md.

Assistant profiles

InControl lets you adjust how the assistant communicates:

Default — professional tone, concise answers, explains when asked.
Minimal — brief answers, minimal explanation, fewer warnings. Good for advanced users.
Detailed — longer answers, proactive explanation, more cautious. Good for learning.

Plugins

InControl has a plugin system: each plugin declares the permissions it needs in a manifest, and runs sandboxed to what it declares. Version 2.0 does not install plugins from the app. The SDK is described in the Reference.

Key settings

The most important settings to know about:

Setting	Where	What it controls
Theme	App settings	Light, Dark, or System theme
Default Model	Inference settings	Which model loads by default
Temperature	Inference settings	Creativity of responses (lower = more focused, higher = more varied)
Context Size	Ollama settings	How much conversation history the model sees (in tokens)
GPU Layers	Ollama settings	How much of the model runs on GPU vs CPU (-1 = all on GPU)

See the Configuration page for the full list.

Troubleshooting for beginners

“No models available” after launch

Ollama needs to be running and have at least one model pulled:

ollama serve          # start the server
ollama pull llama3.2  # download a model

Then restart InControl or re-run the backend check.

App launches but shows a blank screen

A Store install needs nothing else. For a source build, check that the .NET 9 SDK is installed with dotnet --list-sdks.

Responses are very slow

Check that your GPU is being used: ollama ps shows how much of the model is on the GPU.
Try a smaller model (e.g., llama3.2 instead of a 70B variant).
In Ollama settings, make sure GPU Layers is set to -1 (offload everything to GPU).

“Connection refused” errors

Ollama defaults to http://127.0.0.1:11434. Confirm it is running with:

curl http://127.0.0.1:11434/api/tags

If you changed the Ollama port, update the BaseUrl in Inference:Ollama settings to match.

How do I report a bug?

Open InControl and use the “Copy Diagnostics” feature to create a support bundle. This includes logs and system info but never your conversation content.
Open an issue on GitHub and attach the support bundle.