Send a JSON (UTF-8) response with CORS headers. Safe against client disconnects.

Print status with emoji

Synchronous helper. If *event_callback* is supplied, important milestones will be forwarded to that callable as event_callback(ph

Get user input with optional default value.

Utility to send a JSON response with CORS headers.

Utility to send a JSON response with CORS headers.

Update progress with number of items processed

Add a message to a session

Generates a completion, now with optional support for images. Args: model: The name of the generation model (e.g., 'llav

Context manager to time operations

Return one vector *per* span. Args: text: Full document text. chunk_spans: List of (char_start, char_end) offsets for

Get a specific session

Print a formatted section header.

Update progress and send notification

Merge new key/values into an index's metadata JSON column.

Check if Ollama server is running

Get token count for text using the tokenizer.

Get list of available models

Print a formatted header.

Send an SSE event to a specific session

Ensure retrieval pipeline uses the embedding model stored with the first index.

Check if a command exists in PATH.

Send a chat message to Ollama

Factory function to get an instance of the RAG agent based on the specified mode. Args: mode: Configuration mode ("default", "fa

Update session title

Get the correct vector table name for a session by looking up its linked indexes.

Converts a Pillow Image to a base64 string.

Send progress update via Server-Sent Events

Estimate memory usage of chunks in MB

Report final statistics

Generate a session title from the first message

Get all messages for a session

Check if a port is already in use.

Process items in batches with progress tracking Args: items: List of items to process process_func:

Start a single service.

Generator that yields partial *response* strings as they arrive. Example: for tok in client.stream_completion("qwen2", "Hello"):

Validates the model configuration for consistency and availability. Raises: ValueError: If configuration conflicts are detected

Formats the user query with conversation history for context.

Generates a contextual summary using a structured, multi-part prompt.

Ensure a dense MultiVectorRetriever is always available unless explicitly disabled.

Helper to load overviews from a .jsonl file into self.doc_overviews.

Perform the actual conversion using the specified converter.

Print system status summary.

📝 FALLBACK: Simple pattern-based routing (original logic).

Uses a text LLM to synthesize a final answer from extracted facts.

Remove sessions with no messages

Create a new chat session

Ensure required Ollama models are available.

Retrieves all document file paths for a given session.

Get all chat sessions, ordered by most recent

Get database statistics

Aggregate overviews for the given indexes or fall back to global file.

Monitor running services and restart if needed.

Remove an SSE connection

Reranks a list of documents based on their relevance to a query. If *early_exit* is True the cross-encoder scores documents in mini-

Runs the indexing pipeline for the specified documents.

Return appropriate EmbeddingModel implementation for the given name.

Gracefully shutdown all services.

Split one Markdown doc into chunks with max_tokens limit.

Starts the API server.

Convert non-PDF formats using general converter.

Convert PDF with OCR detection logic.

Convert plain text files to markdown by reading content directly.

Computes cosine similarity between two vectors.

Lazily download and load the Provence model exactly once.

Finds a semantically similar query in the cache.