What is DeepSeek AI?
DeepSeek is an AI assistant built on a family of large language models developed by the Chinese AI lab DeepSeek. You can use it to answer questions, write and edit text, summarise documents, translate, brainstorm, analyse data and write code. It's best known for two things: publishing its models with open weights, and showing its full reasoning when it works through a hard problem.
You can use DeepSeek three ways, and this guide focuses on the first:
Chat app
Web, iOS and Android. Free. This guide.
API
For developers. Pay per token. API guide →
Harness Desktop
An agent that works in your files. Harness guide →
Under the hood, the chat app runs on several models. You don't need to choose one to get started - the default (V4.1 Flash) is fast and good at everyday tasks - but knowing that R1 handles reasoning and Coder handles programming explains why the toggles described below exist. See all DeepSeek models compared.
Step 1 - Create an account and sign in
You need a free account to chat. Go to the official DeepSeek website or open the app, and choose one of these sign-in methods:
- Email or phone number - you'll receive a one-time verification code; there's no password to set.
- Google or Apple - one tap on a device already signed in to those accounts.
- WeChat - scan the QR code on the web login page.
Whichever you choose, use the same method every time. Signing in by email one day and Google the next can create two separate accounts with separate chat histories. If your code doesn't arrive, check spam and request a new one; our login guide covers every sign-in problem.
Step 2 - A tour of the interface
After signing in you'll see a clean, two-part layout. Here's what each piece does:
- New chat - starts a fresh conversation. Use it whenever you change topic.
- Chat history - every past conversation, grouped by date. Click to reopen, hover for rename/delete.
- Message box - type here. Shift+Enter adds a line break; Enter sends.
- DeepThink Search toggles - switch on reasoning mode and live web search (explained below).
- Attach - upload documents or images for DeepSeek to read.
- Send - or just press Enter. While a reply is streaming, this becomes a Stop button.
Step 3 - Start your first chat
Click into the message box, type a question in plain language, and press Enter. The reply streams in word by word; you can keep typing a follow-up while it finishes. DeepSeek remembers everything said earlier in the same conversation, so you can refine ("make it shorter", "now in Sinhala", "give me three more") without repeating yourself.
Good first things to try
- "Summarise this in three bullet points: [paste text]"
- "Translate this message into Tamil, keeping it polite: …"
- "I have chicken, rice and spinach. Suggest a 30-minute dinner."
- "Explain [topic] like I'm 12, then like I'm a graduate."
Step 4 - Write prompts that get great answers
A prompt is just your message, but a little structure goes a long way. The most reliable pattern has four parts - task, context, constraints, format:
Prompt templates you can copy
Write a [type of text] for [audience] about [topic]. Tone: [tone]. Length: [n] words. Include [must-have].Summarise the text below in [n] bullet points for someone who [context]. Flag anything that needs a decision.Teach me [topic] step by step. After each step, ask me one question to check I understood before moving on.I'm choosing between [A] and [B] for [goal]. List pros and cons of each, then recommend one and say why.Edit the text below for clarity and grammar. Keep my voice. Show changes as a list after the edited version.Act as a [role, e.g. hiring manager]. Interview me for [job]. Ask one question at a time and give feedback.Five habits that improve every answer
- Give an example of the output you want - a sample paragraph, a sample row of a table.
- Say who it's for. "For my grandmother" and "for a cardiologist" produce very different explanations.
- Ask for a format: table, bullet list, numbered steps, JSON, a 2-column comparison.
- Iterate. The first answer is a draft. "Shorter", "more formal", "add a counter-argument" all work.
- Ask it to ask you. "Before answering, ask me any questions you need" stops guessing on vague tasks.
Step 5 - Use DeepThink (reasoning mode)
Tap DeepThink under the message box and DeepSeek switches to its reasoning model (R1). Before answering, it works through the problem step by step in a collapsible "thinking" panel you can read - and it often catches its own mistakes there.
Use it for
- Maths, logic puzzles and word problems
- Planning with trade-offs (itineraries, budgets, schedules)
- Debugging and tricky code questions
- Analysing arguments or research questions
Skip it for
Quick facts, rewrites, translations and casual chat. Reasoning mode is slower and more verbose - for everyday tasks the default model answers faster and just as well. Read more about when each model wins.
Step 6 - Use web search for up-to-date answers
By default, DeepSeek answers from what it learned during training, which has a cut-off date. Switch on Search and it looks up current web pages first, then answers with numbered citations you can click to verify.
Turn it on for
- News, prices, scores, weather, exchange rates
- Anything that says "latest", "current" or "this year"
- Product specs, opening hours, software versions
- Checking a claim before you repeat it
Step 7 - Upload files and images
Click the paperclip (or drag a file onto the chat) to attach documents and images. DeepSeek reads the content and answers questions about it; you can reference the file in later messages without re-uploading.
What you can upload
| Type | Good uses |
|---|---|
| PDF, DOCX, TXT | Summaries, QA, finding clauses, rewriting sections |
| XLSX, CSV | Explain the data, spot trends, draft formulas |
| Images, screenshots | Read text in a photo, describe a chart, explain an error dialog |
| Code files | Review, explain, find bugs, add comments |
Prompts that work well with files
- "Summarise this in 5 bullets and list any deadlines."
- "Turn the table on page 3 into a clean list."
- "What questions should I ask before signing this?"
- "Compare these two attached quotes side by side."
Step 8 - Voice input
In the mobile app, press and hold the microphone next to the message box and speak; release to send. Your words are transcribed and you can edit them before sending. It supports many languages - speak naturally and switch languages mid-sentence if you need to. On the web, use your device's dictation (Windows key + H on Windows, Fn twice on Mac) to type by voice into the box.
- Speak in full sentences - the transcription is better with context.
- Say "new paragraph" to break up longer dictation.
- Noisy room? Short phrases transcribe more reliably than long ones.
Step 9 - Get coding help
DeepSeek is especially strong at code. Paste a snippet or an error message and describe what you expected. It explains the cause, proposes a fix (ask for it "as a diff" to see exactly what changes), writes tests, converts between languages and documents functions. Code is returned in copyable blocks with syntax highlighting.
Useful prompts
Here's my function and the error. Explain why it fails, then give a minimal fix as a diff.Explain this code line by line as if I'm new to Python. Then suggest one improvement.Write a Python script that reads a CSV of sales, totals by month, and saves a bar chart as PNG. Include error handling.For bigger jobs - refactoring a whole project, running tests, editing many files - move to Harness Coding, where DeepSeek works directly in your repository. Dedicated Coder models are also available through the API for IDE integrations.
Step 10 - Use the DeepSeek mobile app
The app for iOS and Android has everything the web version does - DeepThink, Search, file uploads - plus voice input, share-sheet integration (send text or a photo from another app straight into a chat) and offline access to your history. Chats sync with the web instantly.
- Swipe from the left edge to open your chat history.
- Long-press a reply to copy, share or regenerate it.
- Use the share sheet in Safari/Chrome to summarise a web page.
- Enable notifications if you use long DeepThink sessions.
Step 11 - Manage, share and delete chats
Rename organise
Hover a chat in the sidebar (long-press on mobile) → Rename. Give chats clear names like "Visa letter v2" so the list stays useful. Pin the ones you return to.
Share a conversation
Use the share icon at the top of a chat to create a read-only link. Anyone with the link can view it; remove the link later from the same menu.
Delete privacy
Delete a single chat from its menu, or clear all history in Settings. Deleted chats can't be recovered. Avoid pasting passwords, card numbers or other sensitive data.
10 pro tips for getting more out of DeepSeek
- Set the scene once. Start a chat with "You are my [editor/tutor/coach]. Here's my situation…" and every answer stays in character.
- Ask for options, then pick. "Give me 5 headline ideas" beats "write a headline".
- Request a plan before the work. For long outputs, ask for an outline first and approve it.
- Use tables for comparisons. "Put that in a table with columns X, Y, Z."
- Chain tasks. Draft → critique → revise, all in one chat: "Now critique it as a harsh editor, then rewrite."
- Check facts with Search on. Any date, number or name worth repeating is worth verifying.
- Mind the context window. Very long chats make the model "forget" early details - start fresh and paste a summary.
- Translate with intent. Say the audience and register: "formal Sinhala for a government letter."
- Learn with the Socratic prompt. "Don't give me the answer - ask questions until I get it myself."
- Save your best prompts. Keep a note of templates that worked; reuse them.
Limits, accuracy and what DeepSeek can't do
Knowing the boundaries makes you a better user:
- It can be confidently wrong. Models sometimes invent facts, citations or code that doesn't run. Verify anything important, especially with Search off.
- Knowledge has a cut-off. Without Search, it doesn't know recent events.
- Usage limits. During peak demand you may see "server busy"; wait a minute and retry. See troubleshooting.
- Not a professional. For medical, legal or financial decisions, treat answers as a starting point for a conversation with a qualified person.
- Privacy. Conversations are processed on DeepSeek's servers; read their privacy policy, and don't share secrets. If you need full control, the open-weight models can be self-hosted - see models.
Next steps: beyond the chat app
Choose the right model
V4 Pro, V4.1 Flash, R1 and Coder compared - use cases, inputs and where to access each.
Compare models → APIBuild with the API
Create a key and send your first request in Python, cURL or Node. See pricing.
API guide → HarnessLet it work in your files
Harness Desktop drafts documents, builds spreadsheets, runs scheduled tasks and adds plugins.
Harness guide →Frequently asked questions
Is DeepSeek free to use?
Yes. The chat app on web, iOS and Android is free, including DeepThink and web search. Developers pay per token for the API.
Do I need an account?
Yes - a free account keeps your chat history and syncs it across devices. Sign in with email/phone, Google, Apple or WeChat; see the login guide.
Which languages does DeepSeek support?
It works in dozens of languages, including English, Chinese, Sinhala, Tamil, Hindi, Spanish, French, German, Japanese and Arabic. Just write in your language; you can also ask it to translate between languages.
What's the difference between DeepThink and normal mode?
DeepThink uses the R1 reasoning model and shows its step-by-step thinking before the answer - better for maths, logic and planning but slower. Normal mode uses V4.1 Flash and is faster for everyday tasks.
Can DeepSeek browse the internet?
Only when you switch on Search. It then reads current web pages and cites them. With Search off, it answers from training data with a cut-off date.
Can I upload PDFs and images?
Yes - PDFs, Word documents, spreadsheets, text files, code and images. Attach with the paperclip and ask questions about the content.
Is DeepSeek safe to use for private documents?
Treat it like any cloud service: files are processed on DeepSeek's servers under its privacy policy. Don't upload passwords, card details or confidential data. For full control, run the open-weight models locally.
How is DeepSeek different from ChatGPT?
Both are capable assistants. DeepSeek is free with fewer limits, cheaper via API, publishes open weights and shows full reasoning; ChatGPT has a larger ecosystem of integrations. Full comparison: DeepSeek vs ChatGPT.
Why does it say "server is busy"?
Demand spikes, usually after a new release. Wait a minute, shorten the prompt or turn DeepThink off. Persistent issues are noted on the updates page.
Is this the official DeepSeek site?
No. DeepSeekR1.com is an independent guide and is not affiliated with DeepSeek. See About us and our disclaimer.