Reflacto

ENDE

What Are Local LLMs?

AI assistance that runs on your own computer, not in the cloud. Your reflection stays in your hands.

Desktop only: Local LLMs currently work on desktop computers only. Mobile device support is not yet available due to hardware and software limitations.

Think of it like this

Think of a notebook with a thinking partner built in: you write down a difficult situation, receive fresh perspectives, and the notebook stays on your desk.

You think out loud. The model thinks with you. The conversation stays with you.

It uses technology similar to familiar AI services. The defining difference is not what it does, but where it happens: directly on your computer.

Why Reflacto Uses Local AI

Reflections can contain personal and sensitive details. That is why Reflacto runs AI directly on your device instead of sending your writing to a central cloud service.

What you gain

  • Your reflection content stays on your device
  • After the model download, AI support also works without an internet connection
  • Reflacto does not require an account with a cloud AI provider

What local processing means

  • The model uses storage space on your computer
  • Your hardware determines how fast it responds
  • You complete the one-time setup on a desktop computer

Browser support

Start right away with Chrome’s built-in model

With desktop Chrome 138 or newer, you need neither another application nor a local server. Select it in Reflacto, confirm the one-time model download if prompted, and you’re ready.

  • No extra application
  • No server or CORS setup
  • Ready inside Reflacto
Technical notes for LM Studio, Unsloth Studio, and Ollama

These alternatives run as local servers. Three separate things have to line up before Reflacto can reach them:

1. Reaching your local server from this site

Chrome, Edge and Firefox all allow it. Web standards treat localhost and 127.0.0.1 as trustworthy, so they are not blocked as insecure content. Safari is the exception and still blocks this, so LM Studio, Unsloth Studio, and Ollama will not work there.

2. A permission prompt

Since late 2025, browsers ask your permission before a website may talk to software running on your own computer. Chrome 142 and later, Edge 143 and later, and Firefox 153 and later show a prompt the first time Reflacto reaches your local model. Choose Allow. If you dismissed it, re-enable it in this site’s permissions. A blocked prompt reports itself as a CORS error mentioning a denied address space, so it is easy to mistake for the next point.

3. CORS on your local server

This is the most common cause of failure, and it is a setting on your side rather than anything the browser decides. Local model servers refuse requests from other websites until you tell them otherwise.

  • Ollama: start it with OLLAMA_ORIGINS="https://www.reflacto.com"
  • LM Studio: turn on “Enable CORS” in the Local Server settings, then restart the server
  • Unsloth Studio: keep Studio running on port 8888 and allow Reflacto when your browser asks for local-network access

How It Works

Set it up locally once, then let AI support you directly inside your reflection.

  1. 1

    Choose a local model

    Use Chrome’s built-in model or connect a local provider such as LM Studio, Unsloth Studio, or Ollama.

  2. 2

    Make it available once

    Chrome downloads its model if needed. With an external provider, Reflacto checks the local connection.

  3. 3

    Deepen your reflection

    Ask for another perspective whenever it helps. Your reflection is processed on your own device.

From your first thought to the final suggestion, your reflection stays on your device.

See how Reflacto processes a request

Supported Local LLM Tools

Chrome’s Built-In AI

Nothing to install

Chrome ships a model called Gemini Nano inside the browser itself. There is no application to download and no local server to start, so this is the quickest way to try local AI in Reflacto. Prompts are answered on your device and nothing is sent anywhere.

Requires:Desktop Chrome 138 or newer

Quick Start

  1. 1Open the AI assistant button in the Reflacto header
  2. 2Find the “Chrome’s built-in AI” option at the top
  3. 3If the model is ready, choose “Use Chrome’s built-in AI” and you are done
  4. 4If a one-time download is offered, accept it. The model downloads in the background and only needs to be downloaded once

If Chrome has not switched the model on yet

Chrome ships the built-in model behind flags, so it often needs enabling once. This is a Chrome setting, not a limit of your hardware.

  1. 1Open chrome://flags/#prompt-api-for-gemini-nano and set it to Enabled
  2. 2Open chrome://flags/#optimization-guide-on-device-model and set it to Enabled BypassPerfRequirement
  3. 3Restart Chrome
  4. 4Open chrome://components, find Optimization Guide On Device Model and click Check for update. This downloads the model

LM Studio, Unsloth Studio, and Ollama below need no Chrome flags, so they are the fallback if your device cannot run the built-in model.

LM Studio

User-friendly GUI for running local LLMs with built-in model browser.

Default Port:1234

Download LM Studio

Quick Start

  1. 1Install LM Studio from the link above
  2. 2Download any model in LM Studio
  3. 3Go to “Local Server”, then “Start Server”
  4. 4Enable CORS in server settings
  5. 5Use “Connect to LM Studio” in Reflacto

Unsloth Studio

No-code interface for downloading and running local models through an OpenAI-compatible API.

Default Port:8888

Install Unsloth Studio

Quick Start

  1. 1Install and start Unsloth Studio using the guide above
  2. 2Open http://localhost:8888
  3. 3Download and load a model in Studio Chat
  4. 4Create or copy a local API key in Unsloth Studio
  5. 5Choose “Unsloth Studio” in Reflacto and enter the API key. Reflacto loads your downloaded models automatically

Ollama

Command-line tool for running local LLMs with simple setup.

Default Port:11434

Download Ollama

Quick Start

  1. 1Install Ollama from the link above
  2. 2Set CORS: export OLLAMA_ORIGINS="https://www.reflacto.com,http://localhost:*"
  3. 3Start server: ollama serve
  4. 4Pull a model: ollama pull llama3.2
  5. 5Use “Connect to Ollama” in Reflacto

Questions about local LLM setup? The configuration dialog in Reflacto (accessible via the AI assistant button in the header) provides step-by-step guidance and connection testing to help you get started.