The DeepSeek Wrapper provides a modern, feature-rich chat interface for interacting with DeepSeek AI models.
The interface consists of several key components:
- Chat Area: The main area where conversations are displayed
- Input Box: Where you type your messages
- Settings Button: Access to customize your experience
- File Upload: Button to upload documents for context
- Conversation History: List of your previous conversations
- Command Palette: Press Ctrl+K to access quick actions
- When you first open the application, you'll see an empty chat
- Type your message in the input box at the bottom of the screen
- Press Enter or click the send button to submit your message
- The AI will respond in the chat area
The chat interface supports full Markdown formatting, including:
- Bold, italic, and
strikethroughtext - Code blocks with syntax highlighting
- Lists and tables
- Links and images
Code snippets use lazy-loaded syntax highlighting for faster initial loads. You can:
- Copy code blocks with a single click on the copy button
- See the language identified in the top-right of each code block
Long conversations are easier to scan with automatic date separators. For performance, the UI shows only the most recent portion of the conversation by default. Use the “Load older messages” button to reveal earlier messages. You can configure how many messages remain visible (30/50/100) in Settings → Appearance.
Press Ctrl+K to open the command palette for quick actions such as toggling theme, opening Settings, exporting, or focusing the input.
Under Settings → Appearance you can enable Reduced Motion to disable animations, and toggle timestamps or compact density. The “Thinking…” details panel disables animations when expanded.
To reduce flashing and improve readability while the model is generating a response, the UI now uses a non-flashing "Thinking…" indicator:
- A subtle progress bar communicates activity without strobing effects
- A "Show details" toggle lets you expand a collapsible panel to view the live, streaming content
- When expanded, all animations inside the details area are disabled to avoid visual flicker
- On completion, the details panel auto-expands once, the bar fills, and animations remain off
How to use:
- Send a message as usual
- While the model is thinking, click "Show details" to watch the response stream in real-time
- Click "Hide details" to collapse the panel and keep the conversation compact
Accessibility notes:
- The previous blinking cursor and animated dots have been removed to minimize potential seizure risks
- The progress bar uses a gentle slide animation that is disabled when the details area is expanded
- The indicator and controls are keyboard- and screen-reader-friendly (progressbar, aria-expanded)
To provide context from a document:
- Click the upload button next to the input field
- Select a supported file (PDF, DOCX, or TXT)
- Once uploaded, the content will be processed and included in your next message
- You can reference the document in your query
| File Type | Extensions | Max Size | Notes |
|---|---|---|---|
| PDF Documents | 10 MB | Text extraction from all pages | |
| Microsoft Word | .docx, .doc | 10 MB | Text extraction with basic formatting preserved |
| Plain Text | .txt, .md | 5 MB | Direct inclusion of text content |
After uploading a document, you can reference it in your messages in various ways:
# Examples of document references
"Summarize the uploaded document."
"Extract the key points from the PDF I just shared."
"Answer questions based on the content in the document I uploaded."
"Using the uploaded resume, write a cover letter for a software engineer position."
The document content is automatically analyzed and included in the context for the AI model, making it accessible for the conversation without needing to copy and paste the text manually.
- Documents are processed on the server but are not permanently stored
- Content is extracted and used only for the current session
- X Malware/sensitive content scanning (planned)
- X Automatic expiry after 24 hours (planned)
- Click the settings icon in the top right corner
- In the modal, you can:
- Set your display name
- Choose or upload an avatar
- Configure system prompts
- Adjust UI preferences
| Setting | Description | Default |
|---|---|---|
| Display Name | Name shown in the chat interface | "User" |
| Avatar | Profile picture shown with your messages | Default avatar |
| Theme | Light, Dark, or System-based | System |
| Message Density | Compact or Comfortable view | Comfortable |
| Code Font Size | Size of text in code blocks | Medium |
| Save History | Whether to save conversation history | Enabled |
System prompts help guide the AI's behavior:
- Access system prompt settings via the settings modal
- Enter your custom system prompt or select from templates
- Save your changes
- New conversations will use your custom system prompt
The DeepSeek Wrapper includes several pre-built system prompt templates for common use cases:
- Default Assistant: General-purpose AI assistant
- Code Helper: Specialized for programming assistance
- Writing Assistant: Optimized for content creation and editing
- Data Analyst: Focused on data analysis and interpretation
- Teacher: Designed for educational explanations
- Brainstorming Partner: Optimized for creative ideation
For best results with custom system prompts:
- Be specific about the AI's role and expertise
- Describe the tone and style you prefer
- Specify any limitations or focus areas
- Keep prompts concise but descriptive
- Include examples of desired behavior when possible
Example custom system prompt:
You are a professional software developer with expertise in Python and JavaScript.
Provide detailed, well-structured code examples with clear explanations.
Focus on writing efficient, readable code following best practices.
Include error handling in your examples when appropriate.
The DeepSeek Wrapper includes several AI tools that can be configured through the settings panel:
- Click the settings icon in the top right corner
- Switch to the "Tools" tab
- Configure the available tools:
- Enable or disable specific tools using the toggle switches
- Enter API keys for tools that require external services
- Configure additional settings for each tool (e.g., SMTP server for email)
| Tool | Description | API Key Required |
|---|---|---|
| Web Search | Search the web for real-time information | Yes (SEARCH_API_KEY) |
| Weather | Get current weather and forecasts | Yes (OPENWEATHERMAP_API_KEY) |
| Compose and send emails | Yes (SMTP credentials) | |
| Calculator | Perform calculations | No |
| Date & Time | Get current date and time information | No |
When you enter API keys in the tools configuration panel:
- The keys are saved securely in your local
.envfile - Tools are automatically registered with the appropriate API keys
- The AI can immediately use the configured tools in your conversations
This approach provides several benefits:
- API keys are stored securely on your server, not in the browser
- Configuration is persistent across server restarts
- No need to manually edit configuration files
- Tools can be easily enabled or disabled as needed
Note: The API keys are stored in plaintext in your
.envfile. Ensure this file has appropriate permissions and is included in your.gitignoreto prevent accidental exposure.
Once tools are configured, you can use them in your conversations by asking questions that require their functionality:
- "What's the weather in Tokyo today?" (Weather tool)
- "Search the web for the latest AI research papers" (Web Search tool)
- "Help me draft an email to schedule a meeting" (Email tool)
- "What is the square root of 144?" (Calculator tool)
- "What's today's date?" (Date & Time tool)
The AI will automatically use the appropriate tool when needed to answer your questions.
The DeepSeek Wrapper includes a comprehensive tool status monitoring system and caching mechanism. These features enhance performance and provide better visibility into the tool system:
-
Tool Status Panel: The Tools tab includes a status panel showing:
- Ready, error, and unconfigured tool counts
- Detailed status for each registered tool
- API key validation status
- Last usage information
-
Caching System: Tools use an intelligent caching system to improve performance:
- Reduces redundant API calls for identical requests
- Configurable cache time-to-live (TTL) settings
- Cache hit/miss statistics for performance monitoring
- Automatic cleanup of expired cache entries
The caching system provides several benefits:
- Faster response times for repeated queries
- Reduced API usage and associated costs
- Less network traffic and lower latency
- Protection against rate limiting on external APIs
You can manage the tool caching system directly from the UI:
- View Cache Statistics: See cache size, hit ratios, and history for each tool
- Clear Individual Caches: Each tool has a dedicated cache clear button
- Clear All Caches: Use the "Clear Tool Caches" button to reset all caches at once
- Refresh Tool Status: Update the status display to see the latest information
Note: Clearing the cache doesn't impact tool functionality but may result in slightly slower responses until the cache is rebuilt with new requests.
The caching system uses these default settings:
| Setting | Default Value | Description |
|---|---|---|
| Cache Enabled | Yes | Whether caching is active for each tool |
| Time-to-Live (TTL) | 300 seconds | How long cached results remain valid |
| Cache Key | Request parameters | What determines a cache hit/miss |
These settings provide an optimal balance between performance and freshness of information for most use cases.
The DeepSeek Wrapper will soon support model selection, allowing you to choose from various DeepSeek models with different capabilities:
-
Supported Models: The system will support switching between:
- deepseek-chat: General-purpose chat model (default)
- deepseek-coder: Specialized for code generation and programming tasks
- deepseek-llm-67b-chat: Larger model with enhanced capabilities
- deepseek-llm-7b-chat: Smaller, faster model for simpler tasks
- deepseek-reasoner: Model with enhanced reasoning capabilities
-
Model Settings Panel: A dedicated tab in the settings modal will offer:
- Simple model selection UI
- Model-specific configuration options
- Performance and capability comparisons
-
Model-Specific Features: Special features tailored to each model:
- For reasoning models: Option to show only final answers after processing
- For code models: Enhanced syntax highlighting and code explanations
- For larger models: Warning about potential longer response times
-
Model Indicator: A visual indicator in the UI will show which model is currently active
- Starting a new conversation: Click the "New Chat" button
- Viewing history: See previous conversations in the sidebar
- Continuing a conversation: Click on any previous conversation to resume it
The DeepSeek Wrapper UI supports a range of keyboard shortcuts to improve productivity:
| Shortcut | Action | Description |
|---|---|---|
Enter |
Submit message | Send your message (when input isn't multi-line) |
Ctrl+Enter / Cmd+Enter |
Submit message | Send your message (always works) |
Shift+Enter |
New line | Add a line break in the input area |
Esc |
Cancel | Cancel current input or close modals |
Up Arrow |
Edit last message | Recall and edit your last message |
Ctrl+K / Cmd+K |
Quick actions | Open quick action menu |
Ctrl+/ / Cmd+/ |
Keyboard help | Show keyboard shortcut help |
Ctrl+N / Cmd+N |
New conversation | Start a new chat |
Ctrl+S / Cmd+S |
Save conversation | Save current conversation (if changed) |
Ctrl+F / Cmd+F |
Search | Search within the current conversation |
Ctrl+Backspace / Cmd+Backspace |
Delete word | Delete the word before cursor |
Alt+Left / Alt+Right |
Navigate history | Navigate between conversation history |
Tab |
Auto-complete | When typing code, attempts to auto-complete |
Ctrl+Z / Cmd+Z |
Undo | Undo last edit in input box |
Ctrl+Shift+Z / Cmd+Shift+Z |
Redo | Redo last undone edit |
These shortcuts can significantly enhance your workflow, especially when engaging in extended conversations or coding sessions.
The interface is fully responsive and works on mobile devices:
- Swipe from the left edge to access conversation history on small screens
- All features are accessible through touch-friendly controls