Help & Documentation

Everything you need to know about LinkDok

Your complete guide to saving, organizing, and understanding links — powered by the most advanced AI models. Read top to bottom or jump to any section.

Free • No signup 8 AI Models via NVIDIA NIM AI Playground at /playground 100% local • No data stored

🔗 What is LinkDok?

LinkDok is a smart bookmark manager that goes far beyond saving URLs. It lets you organize your links into categories, automatically extract and understand their content, and then chat with an AI that has actually read those pages — so it can answer questions, explain concepts, and tutor you on anything you've saved.

📁 Save & Organize

Create categories and add any URL. LinkDok will auto-fetch the title, favicon, and description.

🤖 AI That Reads Your Links

Select a category and ask questions. The AI reads the actual page content before answering.

🧠 Multiple AI Models

Choose from 8 powerful AI models: from ultra-fast to deep reasoning, all via NVIDIA NIM.

💬 AI Playground

Free chat mode — chat with any model without needing saved links. Like ChatGPT, but from LinkDok.

⚙️ How It Works

LinkDok uses a multi-step pipeline whenever you ask a question inside a category:

  1. You save links into a category. Any URL works — articles, docs, research papers, YouTube videos, blog posts. LinkDok stores the URL, title, and description.
  2. You open the chat for that category. LinkDok automatically fetches the real content of each link (the text, not just metadata). This step happens once when the chat loads.
  3. You type a question. The AI receives your question plus the extracted content of all links as context. It has actually read the pages.
  4. The AI streams an answer. Response typewrites in real-time. You can see if the model is "thinking" (for reasoning models), then the answer fills in progressively.
  5. You continue the conversation. Follow-up questions remember the entire conversation history, so context builds up naturally.
ℹ️
Content extraction uses Diffbot (when a key is configured) or a direct fetch + text parsing fallback. Some sites block automated access; if content can't be fetched, the AI will say so honestly.

📖 Beginner's Guide

Never used LinkDok before? Here's everything you need in 5 minutes.

Step 1 — Add your first category

Click the + Add Category button (or the + icon in the header). Give it a meaningful name like "React Tutorials" or "Interview Prep". Think of categories as folders for related links.

Step 2 — Add links to the category

Click Add Link inside any category. Paste a URL and hit Enter. LinkDok will fetch the title and description automatically. Add as many links as you like.

Step 3 — Start the AI Chat

Click the chat bubble icon 💬 next to a category name to enter the AI chat for that category. The AI loads content from all your saved links.

Step 4 — Ask anything

Type a question like "What are the main hooks in React?" or "Summarize all these resources" or "Quiz me on this topic". The AI uses your saved links as its knowledge source.

Step 5 — Try the AI Playground

Click the Try [Model] animated button in the left sidebar to access the free AI Playground. No links required — chat freely with any model for general questions, coding, writing, math, etc.

✅
Pro tip: Add 3–6 links per category for best results. Too many links can slow down content extraction; too few may not give the AI enough context.

🎯 Use Cases

Here are real-world ways people use LinkDok:

🎓 Students

  • Save all reference articles for a specific subject into one category
  • Ask the AI to "Create a study guide from these links"
  • Ask it to generate quiz questions before an exam
  • Request simple explanations of complex topics in saved research papers

👨‍💻 Developers

  • Save documentation pages and ask "How do I implement X using this API?"
  • Store Stack Overflow threads and GitHub Issues, then ask for synthesized solutions
  • Use the Playground with Devstral for code generation and debugging
  • Save blog posts about architecture patterns and discuss trade-offs with the AI

📰 Researchers

  • Organize research papers by category (e.g., "LLM Papers", "CV Papers")
  • Ask the AI to "Compare the methodologies across these papers"
  • Request a literature review summary
  • Ask it to identify the key findings and open questions

💼 Professionals

  • Save industry reports and news articles, ask for key takeaways
  • Organize competitor analysis links and chat about strategy
  • Save onboarding docs and use the AI as a knowledge base assistant
  • Collect interview prep resources and request mock questions

✍️ Writers & Creators

  • Save reference articles, ask the AI to help draft content based on them
  • Use Mistral Large (great for writing) in the Playground for creative work
  • Organize inspiration links by theme

🧠 AI Models

LinkDok provides access to 8 specialized AI models via NVIDIA NIM. Each has different strengths. You can switch models at any time using the dropdown in the chat header.

Auto (Smart Routing)
LinkDok's default. Automatically picks the best model based on your question type — coding, writing, reasoning, general knowledge.
✦ ThinkingRecommended
Kimi K2.5
Moonshot AI's flagship model. Exceptional at deep reasoning, mathematics, and multi-step problem solving. Best for hard questions.
✦ Thinking180s timeout
Step Flash
Ultra-fast responses (typically 3–8s). Great for quick questions, simple explanations, and when you need speed over depth.
Fastest
Devstral
Mistral AI's code-specialized model. Best for programming tasks, debugging, code review, and technical explanations.
Code Expert
DeepSeek V3.2
DeepSeek's powerful research model. Excellent at research synthesis, detailed analysis, and long-form content understanding.
✦ ThinkingResearch
Mistral Large
Mistral AI's premier model. Excellent for writing, summarization, long-form content, creative tasks, and nuanced answers.
Writing
Qwen 3.5
Alibaba's versatile model. Powerful all-rounder with strong multilingual support and reasoning capabilities.
✦ ThinkingMultilingual
GLM5
Zhipu AI's general purpose model. Balanced performance across a wide range of tasks.
✦ Thinking

Thinking Mode

Models marked with ✦ Thinking support an extended reasoning phase — the model thinks through the problem step by step before answering. This is visible as a live "thinking bubble" during generation.

Thinking mode has three states, toggled via the pill button inside the model dropdown:

  • Auto — LinkDok decides based on question complexity (default)
  • On — Always use thinking (better accuracy, slower)
  • Off — Skip thinking (faster, good for simple questions)

💬 AI Playground

The AI Playground is LinkDok's free chat mode — like having a personal ChatGPT right inside your workspace. No links required.

How to open the Playground

Click the animated "Try [Model]" button in the left sidebar. A new chat session is created instantly.

What you can do

  • Ask general knowledge questions on any topic
  • Get coding help — write, debug, explain code
  • Creative writing: stories, emails, scripts
  • Math and logic problems (use Kimi K2.5 or DeepSeek V3.2)
  • Translations and language help (Qwen 3.5 is great for this)
  • Brainstorm ideas, get advice, debate topics

Managing sessions

Each playground chat is saved as a session in the left sidebar under AI CHATS. Sessions are auto-titled from your first message. You can:

  • Switch between sessions by clicking them
  • Delete a session using the trash icon on hover
  • Sessions persist in your browser (localStorage)
ℹ️
Playground sessions are stored locally in your browser. Clearing browser data will erase them. There is no cloud sync yet.

⭐ How to Get the Best of LinkDok

Organize your categories thoughtfully

Create focused, topic-specific categories rather than dumping all links into one place. "Python Async Docs" is better than "Programming". Narrower context = better AI answers.

Save quality sources

Official documentation, authoritative articles, and structured content extract much better than paywalled sites, heavy JavaScript SPAs, or PDFs. When a link fails to extract, try an alternative source.

Use descriptive questions

❌ Vague
"Explain this"
"What is it?"
"Help me"

✅ Specific
"Explain the event loop from these Node.js docs"
"Summarize the key findings across these papers"
"Give me 5 quiz questions on React hooks"

Pick the right model for the job

TaskBest model
Quick summary or definitionStep Flash
Deep analysis / researchKimi K2.5 · DeepSeek V3.2
Code writing / debuggingDevstral · Auto
Creative writingMistral Large
General questionsAuto
Multilingual / non-EnglishQwen 3.5

Use conversation history

Don't start a new chat for every follow-up. Continue the same conversation — the AI remembers what was discussed and can refine, expand, or correct its previous answers.

Use Thinking mode for hard problems

For math, logic puzzles, complex coding, or multi-step reasoning — set Thinking to On for the best accuracy. For casual questions, leave it on Auto or Off for speed.

⚠️ Errors & Solutions

ErrorCauseFix
The model took too long to respond Reasoning models (Kimi K2.5, DeepSeek) can take 60–120s on complex questions Try a shorter question, turn Thinking Off, or switch to Step Flash
Rate limit reached You've sent 6+ messages in 60 seconds Wait 60 seconds, then resume. Limits reset automatically.
Could not extract content from link Site blocks bots, requires login, or is a JS-heavy SPA Try a different source for the same topic. Wikipedia, MDN, and official docs extract reliably.
Failed to fetch Network error or the API server is temporarily down Check your internet connection. Try again in 30 seconds.
AI gives wrong or hallucinated info Links didn't extract correctly, or model uncertainty Check if extracted content looks right. Try a reasoning model with Thinking On. Always verify important facts.
Chat history disappears on page reload Messages are stored in memory only (not persisted yet) Export important conversations manually. Persistence is on the roadmap.
Stars / thinking bubble shows but no answer comes Model is still in the reasoning phase (can take 30–90s) Wait patiently — the thinking bubble shows the model is actively working. The answer appears after reasoning completes.
Dropdown cut off on mobile Fixed in the latest version Hard refresh the page (Ctrl+Shift+R / Cmd+Shift+R)
⚠️
If you encounter a persistent error not listed here, try a hard refresh first. Most transient issues resolve automatically.

📊 Rate Limits Explained

LinkDok enforces rate limits to ensure fair usage and keep the service running smoothly for all users. There are two levels:

Client-side Soft Limit

6 messages per 60 seconds per browser session.

If exceeded, you'll see a yellow banner and must wait. The counter resets every 60 seconds automatically.

Server-side Hard Limit

10 requests per 60 seconds per IP address.

Applied to all API calls including content extraction. Returns HTTP 429 with a Retry-After header.

Why are there rate limits?

  • Each AI request consumes NVIDIA NIM API credits
  • Content extraction (Diffbot) has its own usage limits
  • Rate limits prevent abuse and ensure the service stays free and accessible
  • Reasoning models (Kimi K2.5, DeepSeek) use significantly more compute per request

Tips to stay within limits

  • Ask one well-crafted question instead of many small follow-ups
  • Use conversation history — the AI remembers context, so you don't need to re-ask
  • Avoid rapid-fire testing in the Playground
  • Step Flash uses fewer resources and rarely hits limits
ℹ️
Rate limits are per-session (client) or per-IP (server). If you're on a shared network (school, office), the server limit is shared with others on the same IP.

❓ Frequently Asked Questions

Is LinkDok free to use?

Yes. LinkDok is currently free. The underlying AI models are powered by NVIDIA NIM API and there is a rate limit in place to ensure fair usage. There is no subscription or payment required.

Do my links and conversations get stored on a server?

No. All your categories, links, and chat history are stored entirely in your browser's localStorage. Nothing is sent to or stored on any LinkDok server. Your data stays on your device.

The only data that travels to a server is your message and the extracted content of your links — which is sent to the NVIDIA NIM API to generate the AI response, and then discarded.

Why does the AI sometimes give wrong answers?

AI models can hallucinate — they may generate plausible-sounding but incorrect information, especially when the source content didn't extract properly or when asked about something outside the saved links. Always verify important facts from authoritative sources. Using reasoning models (Kimi K2.5, DeepSeek) with Thinking On reduces hallucinations significantly.

Can I save YouTube videos or social media posts?

You can save the URL and the title/thumbnail will be fetched. However, full content extraction from YouTube, Twitter/X, Instagram, and similar platforms is limited or unavailable due to their bot-protection. The AI will still try to answer based on whatever content is accessible, but results will be limited.

How many links can I save per category?

There is no hard limit on the number of links. However, for AI chat performance, we recommend 3–8 links per category. Loading too many links increases the content extraction time and may push the AI response over context limits. For very large collections, consider splitting into sub-categories.

Can I use LinkDok on mobile?

LinkDok is primarily designed for desktop use. A mobile blocker is shown on very small screens. However, tablets (≥768px) and landscape phones work reasonably well with the responsive layout.

What happens if I clear my browser data?

Since all data is in localStorage, clearing browser data (cookies & storage) will erase all your categories, links, and Playground sessions. There is no backup or sync currently. Export important data manually before clearing browser storage.

Why does the AI say it "doesn't have access" to some pages?

Some websites block automated content extraction (e.g., paywalled sites, sites requiring login, aggressively JavaScript-rendered single-page apps). In these cases, the AI will note that the content wasn't accessible and answer from general knowledge instead.

Can I export or import my links?

Export/Import is on the development roadmap. Currently, your data lives only in your browser's localStorage. You can inspect it manually via browser DevTools → Application → Local Storage.

Is this a PWA / can I install it?

Not yet. A Progressive Web App (PWA) version is being considered for future development.

Which browser works best?

LinkDok works on all modern browsers. Chrome and Edge are recommended for best performance. Firefox and Safari are supported. Avoid Internet Explorer — it is not supported.

🛡️ Privacy Policy

Last updated: February 2026

What data we collect

LinkDok collects minimal data:

  • No account registration — no name, email, or password is ever required
  • No analytics — we do not use Google Analytics, Mixpanel, or any tracking
  • No cookies — we do not set any tracking or session cookies
  • No server-side storage — your links, categories, and chat history are stored in your browser only

What is sent to third parties

  • AI messages — your chat messages and extracted link content are sent to NVIDIA NIM API to generate responses. Refer to NVIDIA's privacy policy for their data handling.
  • Content extraction — URLs are sent to Diffbot (if configured) for content extraction. Refer to Diffbot's privacy policy.
  • IP Address — your IP address is visible to our serverless API functions (Vercel) for rate limiting purposes only. It is not logged or stored.

Local storage

All user data (categories, links, Playground sessions) is stored in localStorage in your browser. It is never transmitted to LinkDok servers. You can delete it at any time by clearing site data in your browser settings.

Children's privacy

LinkDok is not directed at children under 13. If you are under 13, please do not use this service.

Changes to this policy

This policy may be updated occasionally. Changes will be reflected on this page with an updated date.

👨‍💻 About the Founder

SK

Sudhanshu Kumar

Founder & Full Stack Developer — LinkDok
BS CSDA — IIT Patna

Sudhanshu Kumar is the Founder and Full Stack Developer of LinkDok, an AI-powered productivity platform integrating state-of-the-art AI models through its AI Playground. He specialises in building scalable serverless full-stack systems, AI-integrated product architectures, and innovation-driven digital platforms. His vision is to evolve LinkDok into a powerful AI-driven productivity ecosystem combining intelligent document tools, AI interaction, and next-generation UX intelligence.

Full Stack Web Development AI Product Architecture Serverless & Vercel UX Intelligence Systems Startup Product Dev NVIDIA NIM
🏆

1st Place — DroidRun DevSprint 2026

Co-creator of DroidScope UX Intelligence with team LastCrusade. A winning AI product built at DroidRun DevSprint 2026 — competing against top engineering teams.

Tech Stack

  • Frontend: React 19 + Vite 7, responsive UI, dark/light mode, animated intro
  • AI Backend: NVIDIA NIM API — 8 cutting-edge models with smart routing
  • Content Extraction: Diffbot API + custom multi-fallback parser
  • Hosting: Vercel (serverless functions, global CDN, 300s timeout)
  • Storage: Browser localStorage — zero backend database, fully private
  • Routing: React Router v7 — /playground as standalone AI chat

Philosophy

  • Privacy-first: No accounts, no tracking, no server-side data storage
  • Open AI access: Multiple powerful models without picking sides
  • Local-first: Your data stays in your browser
  • Honest AI: The AI tells you when it can't extract content or isn't confident
  • Speed matters: Step Flash routes general questions for instant responses

Future Roadmap

  • 📤
    Export / Import — JSON backup of categories and links
  • 🌐
    Cross-device sync — optional cloud backup
  • 📎
    Browser extension — one-click link saving from any page
  • 📱
    PWA support — install LinkDok as a native-like app
  • 🏷️
    Tags & advanced filtering — cross-category organisation
  • 🤝
    Shared workspaces — collaborate on link collections with a team

📋 Changelog

v2.0 — February 2026 Latest

  • AI Playground (free chat mode) with session management
  • 8 AI models via NVIDIA NIM with per-model thinking mode toggle
  • Live thinking bubble during model reasoning phase
  • Smart scroll — no forced scroll during streaming
  • Vercel timeout increase to 300s for reasoning models
  • Mobile UI fixes: horizontal scroll eliminated, dropdown fixed
  • Rate limiting (client 6/60s + server 10/60s)
  • Intro landing page with animated cards

v1.0 — 2024

  • Initial release: save links, organize into categories
  • Basic AI chat using saved link content
  • Dark/light mode
  • Link metadata auto-fetch (title, favicon, description)