Project / About

About JAMES AI

A free, local-first AI assistant that runs entirely in your browser — no account, no server, and no data ever sent anywhere. JAMES combines browser-native inference, streaming responses, and a tool-aware architecture so your conversations stay private while still being powerful.

JAMES preview

What makes JAMES different?

Unlike cloud-based chat apps, JAMES keeps computation local. The model runs in a Web Worker, the UI is responsive, and no conversation data is sent to external servers unless you explicitly request external information through tools.

Local by default. Model weights are downloaded once and cached by the browser. Downloads use chunked fetching so interrupted transfers can recover and cached files load much faster on subsequent starts.

Core values

Privacy first

Threads live only on your device. Clearing site data removes everything: history, session state, and cached model files.

Browser native

No native wrapper is required. JAMES runs in the browser using standard web APIs, no install needed.

Streaming responses

Answers appear token by token, giving a fast and alive experience even when the model computes locally.

Tool aware

Built-in tools let JAMES fetch live weather, lookup Wikipedia, convert currency, and even execute real Python code securely in the browser.

♟️ Play games

Challenge JAMES to a game of Chess or Checkers — rendered right in the chat with full rule enforcement, move highlighting, and AI-powered moves.

Built for the browser

Transformers.js

Handles model loading, tokenization, and generation in the browser, with support for both WebGPU and WASM backends.

Modular Web Workers

Inference runs off the main thread so the UI remains smooth. The worker logic is split into specialized modules for efficiency.

Dynamic Models & Detection

Automatically evaluates your hardware (RAM, WebGPU, WebGL fallback) with advanced heuristics to select the optimal model for your device, including accurate iPadOS detection.

Zero-delay Smalltalk

A fast, built-in heuristic engine instantly handles greetings, jokes, philosophy, and common banter without spinning up the heavy LLM.

Smart UI Modes

Adapts fluidly across devices with dedicated Desktop, Mobile, and 10-foot TV viewing interfaces.

IndexedDB

Stores chat threads and memories locally in the browser using the lightweight `idb` library for robust, encrypted data persistence. No cloud sync.

100% Vanilla JS Architecture

Built entirely without heavy frameworks like React or Vue. Features custom state management, smooth CSS animations, and direct DOM manipulation for maximum speed and zero build step.

Premium Glassmorphic UI

Features rich, dynamically generated CSS gradients, buttery-smooth modal animations, and a polished retro-cyberpunk aesthetic.

Algorithmic Audio Engine

Zero-dependency Web Audio API implementation providing satisfying, dynamic sound cues for interactions without the overhead of heavy audio files.

Modular Tools Bridge

Routes browser APIs like clipboard and geolocation through a safe bridge into a modular tool routing system powered by explicit rules.

Frequently Asked Questions

Is JAMES free?

Yes, completely free. No account, no subscription, no API key. Open the site and start chatting.

Does JAMES send my messages to a server?

No. All AI inference runs locally in your browser using WebGPU or WASM. Your messages never leave your device unless you explicitly use a tool that fetches external data (like weather or Wikipedia).

Does it work offline?

Yes. After the model weights download once (a one-time step), JAMES works fully offline. You can also install it as a PWA for a native-like offline experience.

What browser do I need?

Chrome or Edge for the fastest WebGPU experience. Firefox works via WASM. Safari has limited support. No extensions or plugins needed.

What AI models does JAMES use?

JAMES auto-selects from Qwen2.5 0.5B, SmolLM2 1.7B, Llama 3.2 1B, and Llama 3.2 3B based on your hardware. You can also pick a model manually via the ⚙️ settings panel.

Can I play games with JAMES?

Yes! JAMES has a built-in Chess board (powered by chess.js) and a Checkers board with full rule enforcement — forced captures, multi-jump, and king promotion. Just say "let's play chess" or "start a game of checkers" to begin.

Can JAMES write and run code?

Yes! JAMES has a fully integrated Python environment (Pyodide). You can ask it to write Python code for math, data processing, or algorithms, and it will execute the code directly inside your browser and return the results.

Does JAMES do math and physics?

Yes. Alongside Python, JAMES features native, zero-dependency tools for calculating special relativity, particle physics (like Compton scattering), linear algebra (eigenvalues, Gram-Schmidt QR), and differential equations directly in the browser.

Can JAMES generate art?

Yes, JAMES features a mathematical generative art engine that procedurally generates retro 8-bit symmetrical pixel art (like Space Invaders or GitHub identicons) using pure JavaScript and renders them instantly as SVGs in the chat.

Does JAMES do smalltalk?

Absolutely. JAMES features a built-in smalltalk engine that instantly replies to greetings, jokes, philosophical questions, and fun Easter eggs, bypassing the AI model for zero-delay banter.

Made by

A

SmallTalk Apps

Independent developer · Privacy-first software

SmallTalk Apps builds experiments that prioritize privacy and browser-first execution. JAMES is one such project: an AI terminal that runs locally and keeps your data entirely in your control – no account, no server, no tracking.