Electron AI apps

Problem

You want to add AI features to an Electron app, but you do not want to:

• ship API keys in your app
• pay per request for every user's AI usage
• manage billing for a hosted LLM provider

You also cannot safely expose arbitrary local AI access directly from the renderer process.

Why normal API integration is annoying

Hosted APIs require:

• an API key stored securely (hard in a client app)
• network access on every request
• usage limits, rate limits, and billing
• a backend proxy to hide the key

Direct CLI calls from the renderer are insecure because the renderer can access untrusted web content and should not spawn local processes freely.

How switchboard-ai-sdk solves it

switchboard-ai-sdk is a TypeScript SDK for Node.js and Electron apps that lets developers discover and use local AI tools already installed on a user's machine, including Codex, Claude Code, OpenCode, and Ollama, through one consistent API.

Typical Electron architecture:

• Main process — runs switchboard-ai-sdk and the optional HTTP bridge
• Renderer process — calls the local HTTP bridge through a preload script
• User — brings their own installed AI tools

Minimal code example

Main process:

import { startSwitchboardServer } from "switchboard-ai-sdk";

const server = await startSwitchboardServer({ port: 0 });

mainWindow.webContents.on("did-finish-load", () => {
  mainWindow.webContents.send("switchboard-ready", server.url);
});

Preload bridge:

import { contextBridge, ipcRenderer } from "electron";

contextBridge.exposeInMainWorld("switchboard", {
  getUrl: () => ipcRenderer.invoke("switchboard:url"),
});

Renderer:

const url = await window.switchboard.getUrl();

const res = await fetch(`${url}/chat/ollama`, {
  method: "POST",
  headers: { "content-type": "application/json" },
  body: JSON.stringify({
    messages: [{ role: "user", content: "Hello from Electron!" }],
  }),
});

const data = await res.json();
console.log(data.result.message.content);

Limitations

• Browser-only Electron windows without Node integration cannot use the direct SDK
• You must keep the HTTP bridge on localhost and avoid exposing it to the network
• Agent tools like Codex can read and write files; inform users before connecting