Every model. One tab.
- Status: Active flagship
- Live Demo: shfqrkhn.github.io/ModelTab
- Repository ZIP: Download current main ZIP
- Bundled Tool: AI Studio Cleaner
- Portfolio Role: Primary AI application.
- Maintainer handoff:
docs/AI_MAINTAINER_HANDOFF.md - Repository ZIP policy:
docs/REPO_ZIP_POLICY.md - License: MIT
ModelTab is a no-install, local-first BYOK AI chat PWA for OpenAI-compatible endpoints, local/LAN LLM servers, and native Gemini API keys. It gives users a browser-based alternative to desktop AI chat tools: bring an endpoint, bring a key when needed, and use the provider directly from the browser.
AI Studio Cleaner now lives inside this repository at tools/ai-studio-cleaner/ as an integrated, dependency-free ModelTab utility for cleaning Google AI Studio exports into readable Markdown.
Many AI chat tools require a desktop install, a hosted account, or a single provider. ModelTab keeps the workflow portable: static files, direct provider calls, local settings, and no bundled backend. It can run from GitHub Pages, a self-hosted static folder, or an extracted zip opened through index.html.
- Supports OpenAI-compatible
/chat/completionsendpoints and native Gemini APIs. - Includes presets for OpenAI, Azure OpenAI, OpenRouter, Groq, Gemini, DeepSeek, MiniMax, Mistral, xAI, Together, Perplexity, Fireworks, Cerebras, NVIDIA NIM, DeepInfra, SambaNova, DashScope, Vercel AI Gateway, LM Studio, Ollama, llama.cpp, vLLM, LocalAI, Text Generation WebUI, and LAN endpoints.
- Includes source-dated free/testing presets for local endpoints, Gemini, GroqCloud, Cerebras, Cloudflare Workers AI, and OpenRouter free models without bundling keys or promising free access forever.
- Provides prompt library, system prompt presets, slash prompt search, prompt variables, memory, per-chat context, image input, JSON mode, streaming, regeneration, copy/edit, and tree chat organization.
- Offers encrypted full backup, normal key-free export, import, local wipe, and browser-only operation.
- Includes optional Workspace Agent Mode for explicit selected-folder, read-only inspection with visible tool traces.
- Includes an integrated, local-only AI Studio Cleaner tool for Google AI Studio export cleanup.
-
Open the live demo, self-host the static files, or download the current main repository ZIP and open
index.html. -
Pick a preset for an OpenAI-compatible cloud endpoint, native Gemini, or a local/LAN server.
-
Enter base URL, model, and API key only if the endpoint requires one. Direct browser calls require the selected endpoint to allow browser CORS requests; some cloud providers block static-site browser calls.
-
Use Fetch Models to check the endpoint or Test Prompt to send a tiny 32-token prompt before using a free/testing preset heavily.
-
For LM Studio browser use, start the local server with CORS enabled:
lms server start --cors --port 1234
-
Chat, save prompts, organize conversations, install the PWA on HTTPS/localhost when useful, and export or back up local data.
ModelTab offers convenience presets for providers that currently advertise local, free, or testing-friendly access. These are BYOK presets only: ModelTab never bundles API keys, never enables paid fallback, and treats all cloud prompts as leaving the browser for the selected provider.
Last verified: 2026-07-04.
| Preset | Setup/source |
|---|---|
| Ollama Local | OpenAI compatibility |
| LM Studio Local | OpenAI-compatible endpoints |
| Gemini API | Pricing/free tier |
| GroqCloud | Free plan, rate limits |
| Cerebras | Pricing, OpenAI compatibility |
| Cloudflare Workers AI | Pricing, OpenAI-compatible API |
| OpenRouter Free Router | Free router, pricing/limits |
Free tiers, model names, quotas, retention policies, and browser CORS behavior change outside this repo. Verify the provider account page before relying on any free/testing preset.
- Web/PWA: use the live demo and your browser's install action.
- Local ZIP: download the current main repository ZIP, extract it, and open
index.html. - Self-host: serve this repository root from any static host. No backend or build step is required.
AI Studio export cleanup:
- Open AI Studio Cleaner from the ModelTab sidebar, or directly through
tools/ai-studio-cleaner/. - Drop or select exported Google AI Studio JSON.
- Review, copy, or download the cleaned Markdown, then use Back to ModelTab to return to chat.
- No account, backend, analytics, telemetry, or bundled provider key.
- Keys are session-only unless the user explicitly saves an encrypted local vault.
- Normal export never includes API keys.
- Full backup encrypts keys with a user passphrase.
- Importing a key-free export clears existing session keys and the saved local key vault to prevent stale keys from attaching to restored profiles.
- If saved local data is corrupt, ModelTab restores safe defaults and keeps a local recovery snapshot in that browser.
- Provider content is sent only to the selected endpoint.
- Workspace files are not sent unless the user explicitly attaches files. Workspace Agent Mode can share only visible, successful read-only tool trace snippets from a live selected-folder session; if no verified trace exists, local-file questions fail closed instead of guessing.
- GitHub Pages: serve from the repository root.
- Zip download: extract and open
index.htmldirectly. - PWA shell caching is active on HTTP/HTTPS.
- Local
file://mode skips manifest and service-worker registration for zip-download reliability.
- Full local + live QA gate:
npm run qa - Static regression gate:
npm test - Workspace worker gate:
npm run test:worker - Local-file contract smoke:
npm run test:local-file - Responsive visual gate:
npm run test:visual - Local provider smoke gate:
npm run test:provider - Live Pages gate after deployment:
npm run test:live - Before public ZIP/download-facing changes, also run syntax checks, a secret scan, README link/media checks, local static smoke, provider setup smoke, and GitHub Pages checks.
ModelTab is the main AI flagship. AI Studio Cleaner is maintained as a bundled ModelTab tool, not a separate app surface. Future AI import, cleanup, and migration workflows should extend this repository instead of creating or reviving separate AI utility repos.
.
├── index.html
├── app.js
├── styles.css
├── workspace-worker.js
├── service-worker.js
├── manifest.webmanifest
├── icons/
└── tools/
└── ai-studio-cleaner/
Keep the public runnable tree minimal: only files needed to run the app, bundled tools, README, license material, and GitHub Pages metadata. Never commit API keys, local exports, full backups, or test/runtime artifacts.
MIT. See LICENSE.