Documentation
Setup guides, how-tos, and reference for your secure, fully offline AI assistant.
This flow guides new users through the application for the first time, providing an optional interactive wizard that helps configure essential settings, upload their first model, and understand key features. Users can skip onboarding if they prefer to explore independently.
During the onboarding wizard, users are guided to upload their first AI language model, which is essential for using chat features. This embedded upload experience is simplified and explained within the setup flow.
Choose visual style for chat conversations from multiple preset themes. Each style offers different appearance, message layout, and visual design. Selection determines how chat messages look throughout the application.
Optional onboarding step that helps users upload or create their first dataset for AI-powered document querying. Can upload existing dataset file or start blockify job to create one from documents. This step introduces dataset query features.
Upload documents (PDF, DOCX, TXT, CSV) to chat input to include their content in your message. System extracts text from files and combines with your typed message, allowing AI to reference document content. Up to 10 documents can be added to a single message.
This flow allows users to "pin" important or frequently-used chat conversations to keep them easily accessible at the top of their chat list. Pinned chats appear in a dedicated "Pinned" section, separate from date-organized chats, making them quick to find regardless of when they were last used.
This flow enables users to have conversations with the AI where responses are informed by a specific dataset (corpus) of documents. When dataset query is enabled, the system automatically searches your uploaded documents for relevant information and includes that context in the AI's response, resulting in more accurate, grounded answers based on your specific knowledge base.
This flow allows users to start a fresh conversation with the AI without any predefined templates or settings. It creates a blank slate for free-form conversation where users can ask any questions or have any type of discussion with the AI model.
This flow allows users to create a new chat conversation using a predefined template that includes pre-configured settings, personas, prompts, and behaviors. Templates streamline common workflows like creating social media content, generating proposals, or conducting multi-perspective analysis.
This flow enables users to configure multiple AI "personas" within a single chat conversation. Each persona can have different behaviors, contexts, and purposes - for example, one persona that searches datasets, another that provides creative responses, or multiple personas simulating a team discussion. This allows for sophisticated multi-agent conversations and specialized AI behaviors.
This flow allows users to export their chat conversation as a formatted PDF document, suitable for printing, sharing, or professional documentation. The PDF preserves conversation structure with visual formatting for improved readability.
This flow allows users to export their chat conversation as a plain text file that can be saved, shared, or archived. The exported file contains all messages in the conversation in a readable format, preserving the structure and content for external use.
This flow describes how users continue an ongoing conversation with the AI, sending multiple messages back and forth. The AI maintains context from previous messages in the conversation, allowing for natural, contextual discussions where questions can reference earlier parts of the conversation.
This flow allows users to permanently remove a chat conversation from the application. This is useful for cleaning up unwanted conversations, removing test chats, or managing privacy by deleting sensitive conversations.
This flow enables users to automatically generate a descriptive chat name using AI by analyzing the conversation content. The AI reads recent messages and creates a concise, relevant title that captures the conversation's essence.
This flow allows users to change the name of a chat conversation to something more descriptive and memorable. Custom names make it easier to find specific conversations later, especially when managing many chats.
Choose which embedding model will be used to generate vector representations of processed chunks. Required for creating searchable dataset. Model selection depends on whether creating new dataset or adding to existing one.
Test blockify model with sample text before running full job. Allows verifying model works correctly, understanding output format, and checking processing speed with small test.
This flow describes how users upload document files to a blockify or chunking job for AI processing. The system extracts text from various file formats (PDF, DOCX, TXT, CSV, ZIP) and prepares them for chunking and structuring. Multiple files can be uploaded, and ZIP archives are automatically extracted.
Configure job to start processing automatically at specified future date and time instead of immediately. Useful for scheduling resource-intensive processing during off-hours or specific times.
Choose whether to create new dataset or add to existing dataset when creating blockify/chunking job. Determines where processed results will be stored.
This flow enables users to process documents using AI to break them down into structured "IdeaBlocks" - meaningful chunks of information with names, critical questions, and trusted answers. This structured format makes documents more searchable and useful for AI-powered conversations. The process creates a searchable dataset from your documents.
Create job that splits documents into basic chunks without AI structuring. Faster and simpler than blockify mode, suitable when documents don't need AI-enhanced structure or when speed is priority over sophisticated organization.
Access detailed chunk preview showing exactly how each uploaded file will be split based on current settings. Allows fine-tuning chunk size and overlap while seeing real-time preview of results for each file.
Adjust chunk size and overlap settings during job creation to control how documents are split into pieces. Settings affect search precision, processing speed, and result quality.
Upload AI models (language or embedding models) directly from Settings page through dedicated model management interface. This is the standard location for adding new models outside of onboarding.
Move between different settings categories using tabbed interface to access various configuration areas including models, chat options, benchmarking, and admin overrides.
Configure advanced system-wide performance settings including CPU worker count for parallel processing and default chat template. These settings affect all users and all operations.
Upload pre-processed dataset files directly from Settings page. Alternative access point to dataset upload for users managing datasets through settings interface rather than from job completion context.
Adjust global chat behavior settings that affect how the AI responds, including temperature (creativity), frequency penalty (repetition reduction), lookback size (conversation memory), and dataset query parameters.
Adjust context window size through Settings > Chat Options interface. This is the primary location for context window configuration, accessed through the Settings page tabbed interface rather than other possible access points.
Set a default template that automatically applies when creating new chats by typing without selecting a specific template. Streamlines workflow for users who consistently use same chat configuration.
Navigate through multiple pages of dataset items using pagination controls when dataset contains more items than can be displayed on one page.
This flow enables users to view comprehensive details about a dataset, including all the structured items (IdeaBlocks) it contains, statistics, and metadata. Users can browse through potentially thousands of items, search for specific content, and verify the quality of their dataset before using it in conversations.
View all uploaded datasets in list format with key information like name, status, size, and quick actions. Provides overview of available knowledge bases.
This flow allows users to upload a pre-processed dataset file (in JSONL format) to the application. These datasets contain structured information that can be queried during AI conversations, enabling the AI to provide answers based on your specific documents and data rather than just its general knowledge.
This flow allows users to download their dataset as a JSONL file to their computer. This is useful for backing up datasets, sharing them with others, transferring between systems, or archiving for future use. The exported file contains all structured items with their text and vector embeddings.
Search and filter dataset items to find specific information quickly. Search works across all text fields (block names, questions, answers) with real-time results.
This flow enables users to activate or deactivate a dataset for use in AI conversations. Only one dataset can be active at a time. The active dataset is the one that will be queried when the dataset query feature is enabled in chat conversations, providing context-specific information from your documents.
This flow describes how users monitor background job processing through a persistent status badge that appears in the bottom-right corner of the screen when jobs are running. The badge provides at-a-glance status of processing jobs without requiring users to navigate away from their current task.
View comprehensive list of all processing jobs (past and present) in table format with sorting, filtering, and status information. Provides overview of all document processing activity.
Explore different visualization charts showing job processing performance including progress, speed, latency, and throughput. Charts help understand patterns, identify issues, and assess efficiency.
This flow enables users to access a comprehensive dashboard showing detailed information about a specific processing job, including real-time progress, performance analytics, file-by-file breakdowns, charts, and structured results. This dashboard serves as the central monitoring and analysis interface for document processing jobs.
Observe real-time updates of job processing progress including completion percentage, current stage, processing rate, and estimated time remaining. Enables passive or active monitoring without manual refreshes.
This flow allows users to restart a job that previously failed or was cancelled. The retry attempts to process the job again from the beginning, giving users a chance to succeed after resolving the issue that caused the original failure.
Drill down into specific file within job to see detailed analytics, charts, and processing statistics unique to that file. Useful for comparing file processing performance or troubleshooting specific file issues.
This flow allows users to stop a currently processing job, terminating all active tasks and preventing further processing. This is useful when users need to reconfigure settings, realize they uploaded wrong files, or need to free up system resources.
Filter jobs table to show only jobs with specific status (pending, processing, completed, failed, or cancelled). Helps users quickly find jobs in particular states without scanning entire list.
View detailed information about an uploaded model including its name, type, file path, capabilities, and configuration. Helps users understand model specifications and verify correct model is being used.
Choose which embedding model will be used for converting text into vectors for semantic search. This model is used when creating datasets, querying datasets during chat, and processing documents.
This flow allows users to choose which AI language model will be used for chat conversations. Different models have different capabilities, speeds, and specialties. Selecting the appropriate model ensures optimal performance for your specific needs.
Permanently remove an AI model from the application to free up disk space or clean up unused models.
Upload an embedding model that converts text into numerical vectors for semantic search. Required for dataset query functionality and creating searchable datasets from documents.
This flow allows users to add a new AI language model to the application. Models are required to enable chat functionality. The model file is uploaded from your computer to the application and becomes available for use in conversations.
Adjust how much conversation history the AI model can process at once. Larger context windows allow AI to remember more messages but take longer to process. Finding optimal balance between memory and speed is key to good performance.
Review completed benchmark test results to understand model performance characteristics including speed, latency, throughput, and resource usage across various test scenarios.
Continue previously stopped or interrupted benchmark from where it left off instead of starting over. System automatically skips completed tests and runs only remaining tests.
This flow enables users to run comprehensive performance tests on their AI models to understand how well they perform on their specific hardware. Benchmarking measures speed, throughput, latency, and resource usage across various test scenarios, helping users make informed decisions about model selection and configuration.
Stop currently running benchmark test suite to free up system resources or terminate if no longer needed.
Export all benchmark results as TSV (tab-separated values) file for external analysis, documentation, or sharing. File includes all test results, system information, and performance metrics.