Private desktop client for local Ollama models

Your models.
Your machine.

A private desktop client for local Ollama models. Have useful conversations with models running on your own hardware — without an account, cloud AI, or telemetry.

Try the beta

Fedora / Linux · Beta

A quiet interface

A focused place to talk to the models on your hardware.

Sidebar, model picker, conversation, composer. Nothing else is needed. Nous opens straight into a chat with the models Ollama has installed — no dashboard, no layers of menus, no feeds.

Automatic model discovery
Streamed chat responses
Persistent local history
Markdown and math rendering

Local by design

AI that stays on your machine.

Nous connects to the Ollama server you configure. By default that is 127.0.0.1:11434 — a loopback address on your own machine. Your prompt travels from Nous to Ollama, and the conversation never has to leave your local environment.

The endpoint is configurable, so a remote Ollama server can be used if you choose one. But the default, the recommended path, is your own hardware.

Nousthe client

Ollamalocal runtime

Your modelselected model

Your machinehardware and memory

Everything stays within your local environment

Privacy

Nothing to sign into.

Private by design, not by policy page. There is no account system, no analytics pipeline, and no cloud inference to worry about in the first place.

No account

No account is required to use the application. Open it and start talking to your models.

No telemetry

Nous does not include analytics, tracking, or telemetry of any kind.

No cloud AI

Chat requests go through the Ollama endpoint you configure. Nothing is sent to external servers.

How we handle your data

Conversations persist locally
Settings persist locally
Context is always explicit
No silent capture
Renderer has no filesystem or shell access
No remote scripts, fonts, or images

The window that displays the interface never talks to the network, the filesystem, or the shell directly. Any of that work happens in the application's main process, behind a narrow, validated bridge.

The interface

One window, every state covered.

From the first empty conversation to a history full of past sessions — the same quiet layout throughout.

How does formatting work in replies?

Reading model output

Nous renders assistant replies as formatted markdown, so responses stay readable. Headings, lists, tables, code, equations, and links all work.

Streamed text appears as it is generated.
Code is presented as a copyable card.
Math renders inline with KaTeX.

iℏ \partial /\partialt ψ = Ĥψ

You can also ask for concrete instructions, like an example systemd unit:

ini

[Unit]
Description=Private chat client

[Service]
ExecStart=/usr/bin/nous
Restart=on-failure

Full styling notes live in the beta release notes.

Message Nous…

Local and privateEnter to send · Shift Enter for new line

Readable output

Answers that read well.

Model output is rendered as proper markdown — headings, lists, tables, code, equations, and validated links. Long responses stay navigable; technical answers stay copy-pasteable.

Markdown
Headings
Lists
Tables
Code
Equations
Links

Show me the main Ollama commands with a table and the wave equation for context.

Managing local models

Here are the commands I use most with Ollama:

Command	Purpose
ollama pull <model>	Download a model
ollama run <model>	Start interactive chat
ollama list	Show downloaded models

And the wave equation, in case that helps:

∂²u / ∂t² = c² · ∇²u

For shell use, here is a one-liner equivalent:

bashCopy

ollama run phi4-mini "list the commands as a table"

Full reference: Ollama commands (opens as a validated link in your browser).

＋ Add contextPaste selection…Current windowChoose file…

Current contextinstalling.md — lines 42–68×

Used for this request onlyLocal

Companion

AI that stays beside your work.

A compact, always-on-top window that sits beside your editor, terminal, or browser. It never watches what you do — you choose what it sees.

＋

Add context, explicitly

Attach a pasted selection, the current window, or a chosen file. The source is shown before you send.

⌁

Keep it on top, if you like

Pin the Companion above other windows while you work. Ask without switching away from the task at hand.

••

One request at a time

Context is used for the single request you submit it with — unless you explicitly keep it. Nothing is captured on its own.

Give Nous the context you choose, when you choose it. No passive screen monitoring, no automatic capture.

Models

Bring the models you already use.

Nous discovers every model available through Ollama automatically. Whatever you have pulled, it appears in the picker — ready to chat.

Phi-4 Mini

3.8B

Solid all-rounder for mid-range hardware.

Approx. download~2.5 GB

Llama 3.2

3.2B

Compact, practical everyday choice.

Approx. download~2.0 GB

Qwen2.5

3.1B

Good writer and structured-answer model.

Approx. download~2.0 GB

Gemma 3

12.2B

Capable larger model for stronger hardware.

Approx. download~8.1 GB

Mistral

7.2B

A reliable mid-size general model.

Approx. download~4.4 GB

Llama 3.3

70B

Large model for high-memory machines.

Approx. download~42 GB

Sizes are Q4 quantizations as documented in the project README. Actual footprint varies by architecture and hardware. Not all models perform equally on every machine.

Hardware

Built for the machine you already have.

Local inference runs across a range of hardware. Start small — you don't need a data center to begin.

CPU

Every machine

Works

A good place to start. Small models like TinyLlama and Llama 3.2 1B run on CPU alone.

Integrated GPU / APU

Common on laptops

Better

Mid-size models in the 3–7B range run comfortably with shared memory.

Dedicated GPU

For heavier loads

Largest range

Dedicated VRAM opens up larger models for faster responses.

No performance guarantees implied — real throughput depends on your specific hardware and model choice.

Fedora / Linux

Built for Fedora first.

Nous ships as a native Fedora RPM, with a portable AppImage for other Linux distributions.

Fedoraprimary targetRPMnative packageAppImageportable, any distroLinuxx86_64

InstallFedora

sudo dnf install ./nous-*.rpm   # Fedora (recommended)

InstallAny distro

chmod +x ./Nous-*.AppImage   # portable
./Nous-*.AppImage

Ollama is available directly from Fedora's repositories:sudo dnf install -y ollama

Download the current beta from GitHub Releases ↗

Under the hood

Separated by design.

The interface is a thin presentation layer. It has no direct access to your files, your shell, or the network. Every request travels through a narrow, validated bridge to the main process, which owns all Ollama traffic and local storage.

Renderer — Isolated UI. No Node.js, shell, or filesystem access.
Preload bridge — Exposes only the exact API the window needs.
Validated IPC — Every payload is checked before work begins.
Main process — Owns Ollama, persistence, and window management.
Ollama — Your configured endpoint — local by default.

Renderer

↓

Preload bridge

↓

Validated IPC

↓

Main process

↓

Ollama

No formal security certification is claimed. This is an architectural choice that keeps the interface unprivileged — not a promise against every possible threat.

Philosophy

Quiet software for loud machines.

Nous is deliberately focused. It is a place to talk to the models running on your machine — and nothing else.

No feeds.
No notifications.
No accounts.
No cloud dashboard.
No unnecessary layers.
Just a chat with your models.

The models stay on your hardware. The conversation stays on your hardware. The software stays out of your way.

BetaAvailable today

Nous is currently in beta.

Try the current build today on Fedora or any Linux distribution, or join the waitlist for future releases and updates.

Try the beta

Fedora / Linux · v0.1.0 · MIT License

Your models.
Your machine.

What can I help you explore?

A focused place to talk to the models on your hardware.

AI that stays on your machine.

Nothing to sign into.

How we handle your data

One window, every state covered.

Reading model output

Answers that read well.

Managing local models

Summarize this document in two sentences.

AI that stays beside your work.

Add context, explicitly

Keep it on top, if you like

One request at a time

Bring the models you already use.

Phi-4 Mini

Llama 3.2

Qwen2.5

Gemma 3

Mistral

Llama 3.3

Built for the machine you already have.

CPU

Integrated GPU / APU

Dedicated GPU

Built for Fedora first.

Separated by design.

Quiet software for loud machines.

Nous is currently in beta.

Your models.Your machine.

A focused place to talk to the models on your hardware.

AI that stays on your machine.

Nothing to sign into.

How we handle your data

One window, every state covered.

Answers that read well.

Managing local models

AI that stays beside your work.

Add context, explicitly

Keep it on top, if you like

One request at a time

Bring the models you already use.

Phi-4 Mini

Llama 3.2

Qwen2.5

Gemma 3

Mistral

Llama 3.3

Built for the machine you already have.

CPU

Integrated GPU / APU

Dedicated GPU

Built for Fedora first.

Separated by design.

Quiet software for loud machines.

Nous is currently in beta.

Your models.
Your machine.