Skip to content
YOUR COMPUTER · YOUR MODEL

Local AI deserves
a real workspace.

Connect the browser directly to an OpenAI-compatible model server on this computer. PUI adds projects, files, agents, prompts and organized conversations without proxying the request.

Connect a local modelView setup
Optional local tokenNo pui.ai gatewayLoopback onlyMac and PC
SUPPORTED RUNTIMES

Choose the server that fits your machine.

ORECOMMENDED

Ollama

Simple local model management with an OpenAI-compatible endpoint at 127.0.0.1:11434.

LMDESKTOP UI

LM Studio

Browse and load local models visually, then expose the selected model at 127.0.0.1:1234.

MLXAPPLE SILICON

oMLX

Serve MLX and vision-language models efficiently through 127.0.0.1:8000.

THREE STEPS

From local server to first answer.

  1. 1

    Start the local server

    Run Ollama, start the LM Studio server or launch oMLX with an OpenAI-compatible API. The service stays on this computer.

  2. 2

    Allow the browser origin

    The server must accept requests from https://pui.ai. For Ollama installations that require an explicit origin, start it with OLLAMA_ORIGINS=https://pui.ai ollama serve. If the server requires authentication, store its token in PUI's optional local-provider credential field.

  3. 3

    Connect and discover

    Open Models in PUI, choose the local runtime and run the connection test. In Chromium, approve the Local Network Access prompt when it appears.

Why HTTP works from HTTPS

Current Chromium browsers grant secure sites explicit Local Network Access. PUI declares 127.0.0.1 as loopback, while your local server separately controls CORS. The pui.ai server never participates in that connection.

Already running Ollama?

Open Models and connect to the models on this computer.

Open local-model settings