Glossary

The DocLD REST API provides programmatic access to document upload, parsing, extraction, chat, and knowledge bases.

A secret credential used to authenticate requests to the DocLD API. Passed in the Authorization header.

Processing that runs in the background. Requests return immediately with a job ID; results are available later.

Verifying the identity of the caller when accessing the API. DocLD uses API keys in the Authorization header.

Determining what an authenticated caller is allowed to do. Access to documents, knowledge bases, and resources is scoped by organization and permissions.

Processing multiple documents together in a single operation. DocLD supports batch uploads and bulk extraction for high-volume document workflows.

Coordinates that define a region in a document (e.g., page and rectangle). Used for highlighting and citation of extracted or retrieved content.

A container for objects in object storage. DocLD uses buckets (e.g., on R2 or S3-compatible storage) for document and artifact storage.

A unique identifier for a chunk within a document. Used in metadata, retrieval, and citations to reference the exact passage.

Overlapping content between adjacent chunks. Overlap can improve recall for content that spans chunk boundaries.

The process of splitting documents into smaller segments for embedding and retrieval. DocLD uses semantic chunking to respect logical boundaries.

Source attribution linking answers to document passages. Citations show where information came from for transparency and verification.

The model's generated output in response to an input prompt. Used in chat, extraction, and text generation.

A 0–1 measure of how reliable an extraction or retrieval result is. DocLD provides per-field and overall confidence for extraction.

A minimum confidence score below which results are flagged for review or rejected. Used in extraction and sometimes in RAG.

In PDF, the stream of graphics and text operators that define how a page is drawn. Text extraction reads from content streams.

The maximum amount of input (tokens or text) an LLM can process at once.

A collection of documents used for search, RAG, or analysis. In DocLD, a knowledge base holds a corpus of documents.

A measure of similarity between two vectors based on the angle between them. Used for ranking vector search results.

Exporting extracted or tabular data to CSV format. Useful for spreadsheets, reporting, and downstream systems.

The DocLD web UI for managing documents, knowledge bases, extraction schemas, workflows, API keys, and usage.

The number of dimensions (size) of an embedding vector. Dimensionality affects storage and similarity computation.

Assigning documents to categories or types (e.g., invoice, contract, receipt). Supports routing and schema selection for extraction.

A unique identifier for a document in DocLD. Used in API calls, metadata, and citations to reference the source document.

AI-powered understanding of documents: parsing, extraction, search, and generation. DocLD delivers document intelligence through parsing, RAG, and extraction.

A sequence of processing steps applied to documents, from upload through parsing, chunking, and optional extraction or indexing.

The end-to-end pipeline of ingesting, parsing, and transforming documents into searchable or structured data. DocLD handles upload, parsing, chunking, and extraction.

A defined group of documents (e.g., by filter, folder, or IDs) used for a specific workflow, knowledge base, or extraction job.

Sending a file to DocLD for processing. Upload triggers parsing, chunking, and embedding for RAG and extraction.

A numerical vector representation of text used for semantic search. DocLD uses Pinecone llama-text-embed-v2 for document chunks and queries.

The AI model that converts text into numerical vectors for semantic search.

The high-dimensional vector space where embeddings live. Similarity in this space corresponds to semantic similarity.

A specific URL and HTTP method that exposes an API operation. DocLD endpoints cover upload, parsing, extraction, chat, and more.

A notification that something happened. Webhooks deliver events when documents are processed or jobs complete.

AI-powered extraction of structured data from documents using schemas. DocLD returns field values with confidence scores and citations.

Providing a small number of examples in the prompt to guide model behavior. Reduces the need for fine-tuning.

Defining how source content or document structure maps to schema fields. Used in extraction and export.

The document type and extension DocLD supports for parsing. Supported formats include PDF, images, spreadsheets, presentations, and office documents.

Training a model on additional data to improve performance on a specific task or domain. Contrast with zero-shot and few-shot.

Detecting form fields, labels, and input regions in documents to support extraction of key-value pairs and structured form data.

Keyword-based search over document content. Matches exact or stemmed words rather than semantic meaning.

Verified, human-validated data used to measure extraction accuracy. Ground truth lets you compare AI extraction results against known correct values.

When an LLM invents information not present in its context. RAG and citations reduce hallucination by grounding answers in retrieved document content.

Combining keyword (full-text) search with vector search for better retrieval across diverse queries.

Performing an operation multiple times produces the same result as performing it once. Safe retries rely on idempotency.

The vector index stores embeddings for similarity search. DocLD uses Pinecone as its vector index for document chunks and queries.

Running an LLM to generate output from input. Inference happens during RAG answer generation and extraction.

The process of bringing documents into the system and making them searchable or processable. Includes upload, parsing, chunking, and indexing.

An asynchronous processing unit for parsing or extraction. Jobs track document processing status and return results when complete.

A standard for describing the structure and validation of JSON data. DocLD extraction schemas define field names, types, and optional constraints.

A collection of documents organized for semantic search and RAG chat. Knowledge bases scope which documents are searched when answering questions.

The time from request to response. Lower latency means faster answers and better user experience.

Identifying document structure such as headings, paragraphs, tables, and figures. Layout analysis improves chunking and extraction quality.

Large Language Model. The AI model that powers RAG answers and extraction in DocLD. LLMs generate text from context and instructions.

A lightweight markup language for plain-text documents. DocLD can parse and output Markdown for document content and exports.

Structured information about documents and chunks. Metadata includes file info, page numbers, processing details, and custom tags.

A logical partition in Pinecone for scoping vectors. DocLD uses namespaces to isolate document embeddings and support multi-tenant deployments.

A PDF that contains embedded text and fonts (not just images). Text can be extracted directly without OCR.

Optical Character Recognition converts images and scanned documents into machine-readable text. DocLD uses VLM-based OCR for 50+ languages.

The top-level tenant in DocLD. Resources like documents, knowledge bases, and API keys belong to an organization.

Retrieving a large list in pages (e.g., cursor- or offset-based). API list endpoints return paginated results.

Extracting text, tables, and layout from documents. DocLD parses PDFs, images, spreadsheets, presentations, and more with OCR and semantic structure.

An ISO standard for long-term archiving of PDFs. PDF/A restricts features to improve preservation and compliance.

The vector database DocLD uses for embedding storage and similarity search. Pinecone provides integrated embeddings with llama-text-embed-v2.

Ready-to-use extraction schemas for common document types like invoices, contracts, and resumes. Prebuilt schemas speed up extraction setup.

A time-limited URL that grants temporary access to upload or download from object storage without exposing credentials.

Designing prompts and instructions to get consistent, accurate behavior from LLMs. Used in extraction and RAG.

The practice of crafting instructions and context for LLMs to guide behavior and output quality.

The user's question or search text. Queries are embedded and sent to vector search to retrieve relevant document chunks for RAG.

A limit on usage (e.g., API calls, documents, storage) per plan or organization. Exceeding quota may result in throttling or errors.

Retrieval-Augmented Generation combines document retrieval with AI generation for accurate, citation-backed answers from your documents.

A cap on the number of requests or operations per time period. Rate limits prevent overloading the API.

Re-running ingestion (parse, chunk, embed, upsert) on already-uploaded documents. Used after changing chunking or embedding settings.

Reordering retrieval results by relevance before generation. DocLD supports heuristic, LLM, and hybrid reranking for improved RAG accuracy.

The step of fetching relevant documents or chunks for a query. In RAG, retrieval precedes generation.

A document stored as images of pages (e.g., from a scanner). Text must be extracted using OCR.

A JSON schema that defines which fields to extract from documents. Schemas specify field names, types, and instructions for AI extraction.

Software development kit—libraries and tools for integrating with the DocLD API in your preferred language.

Search by meaning rather than exact keywords. Uses embeddings and vector similarity to find relevant content.

A chat conversation with history. Sessions preserve context across messages for multi-turn RAG chat.

A measure of how similar two embedding vectors are. Higher scores indicate greater semantic closeness.

The original file or document uploaded for processing. Parsing and extraction operate on the source document.

Data in a fixed schema (e.g., JSON, CSV) with named fields and types. Extraction produces structured data from unstructured documents.

Instructions that set the LLM's role, behavior, and constraints. Used in chat and RAG to enforce citation and tone.

Identifying and extracting tabular data from documents as structured rows and columns. Supports RAG and schema-based extraction.

An LLM parameter that controls randomness of generated output. Lower values are more deterministic, higher more diverse.

A logical boundary for multi-tenancy. In DocLD, the organization is the primary tenant; resources and billing are isolated per tenant.

Pulling raw text from documents regardless of format. In PDFs and images, this may involve parsing content streams or running OCR.

The rate at which documents or operations are processed. Higher throughput means more documents per unit time.

Splitting text into tokens for LLMs and embedding models. Token boundaries affect chunk size and context limits.

The number of chunks retrieved from vector search before reranking or generation. Top-k balances recall with context size and latency.

A sampling parameter that limits the model to the smallest set of tokens whose cumulative probability exceeds p. Controls diversity of output.

What starts a workflow. Triggers include upload, webhook, schedule, or manual run.

Raw documents (PDFs, images, text) without fixed schema. DocLD parses unstructured data into structured content for extraction and search.

A numerical representation of text or data. Embeddings produce vectors for semantic similarity and search.

A database optimized for storing and querying vector embeddings. Enables fast similarity search for RAG and semantic search.

Similarity search over embeddings to find documents or passages that match a query by meaning. DocLD uses Pinecone for vector search.

HTTP callbacks that notify your system when events occur. DocLD webhooks fire on document processing, extraction completion, and workflow events.

The JSON body sent to your webhook URL when an event occurs. Contains event type, resource IDs, and relevant data.

Automated document processing pipelines with triggers and integrations. Workflows run parse, extract, or other steps when events occur.

A single execution of a workflow. Each run has a status, logs, and output from the workflow steps.

Extensible Markup Language used for structured output. DocLD can return extraction results in XML format for integration with downstream systems.

Streaming or incremental delivery of results. DocLD can yield extraction results incrementally as they are produced, reducing latency for large documents.

Extracting structured data from documents without training on your specific documents. DocLD uses zero-shot extraction with schema-defined instructions.