Windows setup guide

Private AI character chat on Windows

You need a Windows PC, Murmur Sphere, Ollama, and a compatible AI model. Here’s what each part does, what stays on your computer, and where the internet is still involved.

See how it fits together

How it fits together

Three pieces with different jobs

These are separate downloads. You choose the model and run it on your own PC.

01

Murmur Sphere

The free character-chat workspace. It manages characters, optional Persona context, and conversations. It does not require an account.

02

Ollama

The separate local service that runs a compatible AI model. Keep it open when you want to chat.

03

A local model

The model generates replies. Models vary in download size, memory use, and speed; choose one that fits your computer rather than assuming the biggest is best.

Before downloading

Check your PC first

  • Windows: the current public installer requires Windows 10 22H2 x64. See the Download page for the current requirement and installer details.
  • Storage and memory: the app and your model need room. Model requirements vary, so review the sizes and hardware guidance before choosing one.
  • Internet for setup: you need it to download the app, Ollama, and a model. Local chat prompts and replies do not need a hosted AI account.
  • Installer warning: this early alpha is unsigned. Windows may show “Unknown publisher”; download only from this site and keep Windows Security on.

How to choose and load a model

First conversation

From installs to a reply

  1. 1

    Install Murmur Sphere from the official Download page and create a local profile.

  2. 2

    Install and open Ollama. In Murmur Sphere, choose a compatible model; download it if needed, then select Load.

  3. 3

    Wait for the model status to say Ready. Selecting a model is not the same as loading it.

  4. 4

    Create a character or import a supported card, review its contents, and send a message. A Persona is optional.

The first-chat walkthrough and character-card guide have the detailed in-app steps.

Model LibrarySynthetic capture
Synthetic Murmur Sphere Model Library showing installed and download choices.
Synthetic capture of the model choices; the model must be loaded and Ready before chatting.

Privacy, plainly

What “private” covers

On your computer

Character prompts and replies go to the local model through Ollama. Murmur Sphere does not silently switch to a hosted AI. Current saved profiles are local plaintext with integrity checks, so protect your Windows account and device; local does not mean encrypted at rest.

Other connections

Website visits use Cloudflare Web Analytics. Chosen downloads, a small startup release check, optional public images, email, and creator support can use the internet. The app does not send your chats to website analytics. Read the privacy details for the full boundary.

If something stalls

A few common snags

  • No Ready status? Check that Ollama is running, select the exact installed model, and wait for Load to finish. See troubleshooting.
  • Model too slow or cannot load? Try a smaller compatible choice that fits your PC. Size and speed depend on the model and hardware; see local-model help.
  • Windows blocks the installer? Check the official filename and the unsigned-installer instructions on the Download page. The optional file check compares its SHA-256 to the official record; it is not a malware scan.

This is an early alpha. The known issues page has the current limits.

Ready when you are

Start with the setup that fits your PC.

Local chat gives you control over the model and your files, with the tradeoff of installing and maintaining the parts yourself.