BotConnector
ID

BotConnector Local

Run AI models on your own computer.

One command starts a private chat page on your machine. The model runs on your CPU or GPU, keeps working without internet, and can also answer chats you send from the BotConnector web app.

npx https://app.botconnector.id/device-cli-launcher-v1.0.2.tgz offline --allow-local-ai

Needs Node.js 22.12 or newer. Nothing else to install. Which models fit my computer?

Two ways to use the same model

The model always runs on your computer. Only the window you talk to it through changes.

BotConnector Local chat page in the browser.

Offline, in your browser

A chat page at 127.0.0.1 with chat history, file attachments and model management. Works with the network cable pulled out.

BotConnector web app using a model on the user's laptop.

From the web app

Pair the device with a code from the web app and pick Local. Your messages go to your computer; the answer comes back to the web app.

Use Local AI with Switchboard, too

In addition to offline chat and the existing WebApp pairing above, models running through the original BotConnector Local Core can power Switchboard v0.4.2 agents on the same computer. This does not replace WebApp device pairing.

  1. Install or update Switchboard with npm install -g @botconnector/switchboard@latest.
  2. Open the bundled original model manager using sbx local. Explicitly select and start a model; check status with sbx local status.
  3. Run sbx web and choose Settings → Models → Local AI to use a running model.

Switchboard uses your own local hardware, not the website server's GPU. GGUF model weights and runtimes are not downloaded automatically. If your local AI stops, Switchboard does not silently send your request to Cloud.

Explore Switchboard Local AI · v0.4.2 release / Windows portable

Getting started

  1. Open a terminal (PowerShell on Windows) and run the command above. A chat page opens in your browser.
  2. Open Models. BotConnector checks your RAM and GPU and marks each model great, ok or warn for this machine. Download one.
  3. Chat. To use the model from the web app, run the command with connect instead of offline and enter a pairing code from the web app. How to get the code.

Which engine runs the model

BotConnector uses what you already have:

  • Lemonade or Ollama, if one is running.
  • Otherwise it downloads the official llama.cpp build for Windows or Linux — the GPU build when it finds a GPU, the CPU build when it does not.
  • On macOS, install Ollama first; BotConnector will use it.

Connect your computer to the web app

Pair once with a code. After that the device reconnects by itself every time BotConnector Local runs.

  1. Sign in at app.botconnector.id. In the left bar, click the laptop icon to open Devices.
  2. Click Generate pairing code. The code is valid for a few minutes; the expiry time is shown under it.
  3. Enter the code on your computer, in one of two places:
    • in the terminal: run the command below and type the code at Pairing code:
    • in the BotConnector Local page: Connect Web App → paste the code → Connect
    npx https://app.botconnector.id/device-cli-launcher-v1.0.2.tgz connect --allow-local-ai
  4. The device appears under Your devices as Online. In the chat, switch from Cloud to Local and pick a model from your computer.

To stop, close the terminal or click Disconnect. Pairing is saved only for your user account on that computer.

Devices panel in the web app: Generate pairing code, a pairing code with its expiry time, and a paired Windows laptop shown as Online.

Device fit: models chosen by your real hardware

Device fit in BotConnector Local reads this computer's processor, memory, graphics and accelerator, and the engines you have (llama.cpp, Ollama, Lemonade). It ranks models for what you want to do, then measures them on your machine after you download.

Device fit, Recommended tab: this device (AMD Ryzen AI 7 350, 31.1 GB, Radeon 860M, NPU), use cases, Fast/Balanced/Quality, and a recommended model with download size and RAM headroom.

Before you download: Recommended

Pick use cases (General, Bahasa Indonesia, Coding, Reasoning, Vision, Tools) and Fast, Balanced or Quality. Each model shows its download size, the RAM it needs and the headroom left after loading, marked Estimated.

Device fit, Installed tab: Qwen 2.5 1.8B measured at 464.4 tokens per second prompt and 18.5 generate; lfm2.5 8b at 21.8 tokens per second.

After you download: Installed

Benchmark one model or all of them on this machine: prompt and generation speed, load time and end-to-end time, marked Measured. Tick Compare to see two side by side.

Screenshots from a laptop with an AMD Ryzen AI 7 350 and 31 GB of RAM. Open Device fit from the sidebar after starting BotConnector Local; the same device and models are available in the web app once you pair it.

Local app, offline

npx https://app.botconnector.id/device-cli-launcher-v1.0.2.tgz offline --allow-local-ai

Web app, paired with this device

npx https://app.botconnector.id/device-cli-launcher-v1.0.2.tgz connect --allow-local-ai

Stays on your device

The chat page only listens on 127.0.0.1 and needs a random session token. Messages and files are not sent to BotConnector.

Nothing runs without you

Model downloads and inference need --allow-local-ai. Remote tools stay off unless you allow them for that session.

Closing it stops it

The pairing lives only in the running process. Close the terminal or press Ctrl+C and the device goes offline.