Paperwise

##Functional characteristics - ** Document upload and management ** -supports PDF, TXT, MD, DOCX, DOC formats, drag-and-drop uploads - **OCR Optical Character Recognition ** -Supports LLM OCR (Multimodal Model) and local Tesseract dual modes, which can be automatically switched - ˇ ** Automatic extraction of metadata ** - AI automatically identifies title, date, type, label, and communication party - **Ask Your Docs** -Grounded Q&A based on document content and supports traceability - ** Keyword search ** -Full text search for all documents (English only) - **Collections Document Set ** -Organize documents into collections for group Q & A - ** File deduplication ** -Automatic prevention of duplicate uploads based on SHA256 checksum - ** Multi-theme switching ** - Atlas / Ledger / Moss / Ember four themes - ** Multi-user system ** -User registration, login, password modification, model configuration is stored independently by user ˇ* * Important **: Please use the ** English document ** to test the search and Q & A functions. Metadata can be uploaded and extracted from Chinese documents, but search and Q & A do not support Chinese content (original project limit). ###Registered account password Paperwise does not have a visitor mode, and you must create an account for the first time: 1. Click the **Sign Up** tab on the login page 2. Fill in Full Name, Email, Password (more than 6 digits recommended) 3. Click **Create Account** to complete the registration 4. Automatic login to enter the main interface ###Configure AI model (required for first use) The metadata extraction, OCR, and Q & A functions after document uploading all need to be configured with AI model connection. Each user's model configuration is stored independently. 1. Click on the left navigation bar **Settings** 2. Click on the top **Model Config* sub-tab 3. Click **Add Connection** in the **Model Connections** area 4. Select Provider (OpenAI / Gemini / Custom) 5. Fill in API Key, Base URL (Custom type is required), Model 6. After saving, assign models to the following three tasks in **Task Assignments**: - **Metadata Extraction** -Metadata extraction (required, otherwise it cannot be uploaded) - **Grounded Q&A** -Document Q & A (required, otherwise Ask Your Docs will not be available) - **OCR** -Optical Character Recognition (select LLM or Local Tesseract) >** Quick configuration **: Just add a Provider connection and use the same connection for all three tasks to start using it. If most of the documents are clear text PDFs, you can use a faster model first, and then replace it with a stronger model if the results are not satisfactory. **OCR mode description **: | mode| description| applicable scenarios| |------|------|----------| | LLM OCR |Send page pictures to multimodal model processing| Scanners, forms, image-intensive PDFs| | Local Tesseract |Using tesseract + pdftoppm local processing within a container| Privacy sensitive scenes, clear printed scans| | Auto Switch| Automatic fallback to OCR when direct text extraction is not effective| Mixed document scenarios| ** Official recommended model configuration: ** Paperwise supports both GPT and Gemini model systems, and it is recommended to configure them independently according to tasks to achieve the best results: | task| GPT recommended| Gemini recommended| description| |------|----------|-------------|------| | OCR | `gpt-5-mini` | `gemini-2.5-flash` |Fast multimodal models for scans and forms| | Metadata Extraction | `gpt-5-mini` | `gemini-2.5-flash` |Balanced options for structured field extraction| | Grounded Q&A | `gpt-5.1` | `gemini-2.5-pro` |Cross-document Q & A scenarios suggest using a stronger reasoning model| | Lightweight classification/batch processing| `gpt-5-nano` | `gemini-2.5-flash-lite` |Ideal for lightweight classification and diversion tasks| >If most of the documents are clear text PDFs, you can use a faster model (mini/flash) first, and then replace it with a stronger model if the results are not satisfied. ###Upload a document 1. Click **Upload** on the left to enter the upload page 2. Drag and drop or click to select a file (PDF/TXT/MD/DOCX/DOC supported) 3. Click **Upload Selected Files** 4. Go to the **Processing Queue** to check the processing progress 5. Waiting status changes to **Ready** >Paperwise will first try to extract text directly from the document. If it is a scanned copy or PDF image, it will be processed according to the OCR mode you set. >ˇ The same file (same SHA256 checksum) will not be uploaded repeatedly. ###Searching for documents > ˇ* * Important restrictions **: Keyword Search function ** only supports English keywords ** search, and Chinese keywords will return null results (original project restrictions). The search matches the document ** body content **(OCR extracted text) and does not search for titles or tags. | step| operation| test data| expected results| |------|------|----------|----------| | 1 |Click on the left **Search**| - |Display keyword search interface| | 2 |Enter ** English ** search keywords| `agile` or `risk management`| Returns matching results| | 3 |Clear the search box| - |Clear results| ### Ask Your Docs > ˇ Model configuration needs to be completed first (Grounded Q&A tasks have been assigned), and there is at least one processed ** English ** document️ | step| operation| test data| expected results| |------|------|----------|----------| | 1 |Click on the left **Ask Your Docs**| - |Display the Q & A interface| | 2 |Enter English questions and submit| `What are the four steps of risk management?` | AI returns the answer and displays the reference source| | 3 |View Citations form| - |Display source documents and reference fragments|