Chat Assistant Box

A fast, modern AI chat interface with streaming replies, markdown + code highlighting, a dark/light theme, and a Cloudflare Worker backend that talks to a lineup of fast "Flash-tier" models across providers. Vanilla JS, no frontend build step, PWA on Android.

Live

• Web app: https://chat.uft1.com
• Google Play: Android TWA, package com.uft1.chat.twa
• Legacy vanity hosts redirect to the current domain via Cloudflare Redirect Rules
• More projects: https://jovylle.com

Models

Streaming responses across OpenAI-compatible providers — pick from the model selector:

Model	Provider
GPT-4o Mini (default)	OpenAI
Gemini 2.5 Flash	Google
DeepSeek V4 Flash	DeepSeek
Qwen Flash	Alibaba (DashScope)
GLM 4.6	Z.ai

A model whose server-side key isn't configured is hidden/disabled automatically. Bring your own key for any provider to call it directly from the browser.

Features

• Streaming chat – Tokens render progressively as the model generates them
• Multi-provider – Fast Flash-tier models across OpenAI, Google, DeepSeek, Qwen, GLM
• Markdown & code – Replies rendered with marked.js + Prism.js highlighting and copy buttons
• Instant write – The input is focused and typeable before any network call
• Drafts per conversation – Inputs auto-saved locally per chat
• Quick replies – Suggested follow-up prompts ("Explain simpler", "Give examples", …)
• Message edit & resend – Edit any user message and regenerate from that point
• Regenerate / stop – Regenerate the last reply or cancel mid-stream (partial text kept)
• Read aloud – Listen to any reply via the Web Speech API
• Conversation management – Rename, pin, delete, search via the sidebar
• Export & import – Download chat history (JSON/TXT/MD) and restore backups
• Custom API key – Use your own provider key; it stays on-device, never sent to our server
• PWA-ready – Manifest + service worker + install prompt; offline app shell
• Accessibility – ARIA labels/roles, keyboard support (Escape clears input)

Setup

1. Clone and install

   git clone https://github.com/jovylle/chat.git
   cd chat
   npm install
   

2. Provider API keys — set the ones you want enabled (see .env.example). Locally: a .dev.vars file (gitignored). In production: wrangler secret put <NAME>.

   Variable	Provider
   MY_OPENAI_API	OpenAI
   MY_GEMINI_API	Google Gemini
   MY_DEEPSEEK_API	DeepSeek
   MY_QWEN_API	Qwen / DashScope
   MY_GLM_API	GLM / Z.ai
   SESSION_SECRET	HMAC pepper for accounts/sessions

3. Run locally (Wrangler)

   npm run dev            # wrangler dev → http://localhost:8787
   

4. Deploy — npm run deploy (wrangler deploy) uploads the Worker + static assets. First time: wrangler d1 create chat-assistant-box (paste database_id into wrangler.jsonc), wrangler secret put … for each key, then npm run db:migrate:remote. Point chat.uft1.com at the Worker (Custom Domain).

Using your own key

Open the gear icon → Use Your Own API Key, pick a provider, and paste your key. The browser then calls that provider directly (OpenAI-compatible /chat/completions), and the key stays in localStorage — our Worker proxy never sees it. Clear all data in settings to rotate it.

Cost / abuse notes

There is no per-request rate limiter by design. The real safety net is a spend cap set in each provider's own dashboard (your keys, your limits). The Worker applies only a generous input-size guard against obviously abusive payloads.

Android app (AAB for Google Play)

The app is a PWA packaged as an Android TWA. Live URL for PWA Builder / Bubblewrap: https://chat.uft1.com. Package ID: com.uft1.chat.twa (matches public/.well-known/assetlinks.json). Signed AABs live in releases/. Because the TWA just loads the live site, a new AAB is only needed for manifest/package changes — not for frontend/backend updates.

Project structure

• public/index.html – Static SPA + all inline CSS/JS + the MODELS registry
• worker/index.js – Worker entrypoint + /api/* router (static via ASSETS.fetch)
• worker/chat.js – streaming proxy (PROVIDERS + MODELS registries, fetch + tee())
• wrangler.jsonc – Worker config (static assets, D1/R2 bindings, migrations dir)
• public/_headers / public/_redirects – security/cache headers, redirect notes
• migrations/ – D1 SQL migrations (accounts/conversations, Phase 2+)
• AGENTS.md / CONTRIBUTING.md / CHANGELOG.md – docs

License

MIT · jovylle