Chatternel 🖤

A private, mobile-first chat client for OpenRouter. Bring your own API key, talk to any model, keep every word on your own device.

Try it → chatternel.vercel.app — no signup, no install, no waiting.

---

Chatternel is a single-page web app that talks straight to OpenRouter's API from your browser. There is no backend, no account system, and no sync server — nothing between you and the model that could log, moderate, or lose your conversations. Your API key and chat history live in your browser's own storage and leave your device only if you export them yourself.

Add it to your phone's home screen and it behaves like a native app. Open it on a desktop and it grows into a full two-pane chat client.

Why Chatternel?

Most ways to use OpenRouter come with a catch:

• OpenRouter's own chat is heavy, slows down with long conversations, and layers extra moderation on top of the model.
• Self-hosted frontends give you control, but you have to actually host them — and most are rough on mobile.
• Third-party clients are convenient… and store your conversations on their servers.

Chatternel is the middle path: zero install, zero backend, genuinely good on a phone, and private by architecture rather than by promise.

Features

🎛️ Total model control

• Any model, by ID — type any OpenRouter model ID directly; you're never limited to a curated dropdown.
• Provider pinning — route requests to a specific provider (DeepInfra, Together, Fireworks…), with your choice of strict failure or automatic fallback.
• Sampling control — temperature and top-p sliders, sent only when you've actually changed them.

🧠 Built for reasoning models

• Set reasoning effort (auto / off / minimal → max) or an exact token budget.
• Reasoning traces stream in above the reply, collapsible with one tap — including <think>-style traces from models that inline them.

🖼️ Show it, don't describe it

Tap the + next to the send button to attach a picture — a screenshot, a photo, a diagram, a page of handwriting — and ask about it in the same message.

• Attach however you like — pick files with +, paste straight from your clipboard, or drag them onto the composer.
• Several at once — send up to 8 images per message; they show as thumbnails you can remove before sending, and stay with the message afterwards — tap one to see it full size.
• Changed your mind? — editing one of your messages lets you swap, add or drop its pictures too, then send it again.
• Kept light — big pictures are resized in your browser before they're sent, so requests stay quick and your saved chats stay small.
• PNG, JPEG, WebP and GIF, sent to OpenRouter exactly the way its API expects. Just make sure the model you've picked can actually see — a text-only model will simply reject the request.

🫥 Hidden blocks

Elaborate system prompts often make the model emit bookkeeping you don't want to read but do want the model to keep — status headers, tracking notes, scratch metadata, and the like. Define your own marker pairs in Settings, and any matched region folds away into a collapsible block, just like reasoning. It's display-only: the text stays in the message and is still sent to the model on the next turn — only your on-screen view is tidied.

⚡ Streaming that doesn't flinch

Replies render live, but finished paragraphs and code blocks never re-flow or flicker mid-stream — even in very long conversations. Your question pins to the top of the view while the answer fills in below.

✏️ Every message is editable

• Edit any of your messages and regenerate the conversation from that point.
• Edit the model's replies too — the conversation history is yours to shape.
• Redo a reply and flip between alternative versions (‹ 2/3 ›).
• Continue a reply that got cut off with one tap on an empty input.
• Copy any message, or any code block, in one tap.

✨ Summarize instead of forgetting

Long conversation eating your context window? Generate an AI summary of it, review and edit it yourself, choose how many recent messages to keep, and carry on. The summary rides along invisibly as context in future turns — and feeds into the next summary when you compress again.

📝 Proper rendering

Markdown, headings, lists, quotes — and fenced code blocks with syntax highlighting in ~35 languages, each with a copy button.

🔒 Private by architecture

• Everything stored in your browser's own database, on your device, full stop — with room for thousands of messages and plenty of pictures.
• One-tap JSON export/import for backups or moving between devices.
• No telemetry, no analytics, no cookies, no external requests — even the fonts are bundled.
• No moderation or rewriting layer: what the model says is what you see.

📱 Mobile-first, desktop-ready

• Slide-over chat panel, swipe-back gesture, safe-area aware, installable as a PWA.
• On desktop: persistent sidebar, hover states, Enter to send / Shift+Enter for a new line.
• Focus rings, keyboard navigation, and reduced-motion support included.

Getting started

1. Grab an API key at openrouter.ai
2. Open chatternel.vercel.app
3. Paste the key in Settings, pick a model, start chatting

That's it. Your key is stored locally and sent only to OpenRouter, with every request visible in your browser's network tab.

Development

React 19 + TypeScript + Vite. Fully client-side — deploy the dist/ folder to any static host.

npm install
npm run dev      # local dev server
npm run build    # production build → dist/
npm run preview  # serve the production build

License & disclaimer

Chatternel is an independent project, not affiliated with OpenRouter. You bring your own API key and pay OpenRouter directly for what you use.