com.mcparmory/ragie

Create white-labeled connector credentials for integrating with third-party services. This establishes authentication configuration that enables secure API access to supported providers.

Establish a connector for a specified authenticator with provider-specific credentials (e.g., Google Drive refresh token). Configure document processing options like OCR mode, media handling, and metadata.

Create a new connection for non-OAuth data sources such as S3-compatible storage, Freshdesk, or Zendesk. Configure the connection with source-specific settings and optional processing parameters.

Upload and ingest a document for processing and retrieval. The document progresses through multiple processing stages (pending → partitioning → indexed → ready) and becomes available for retrieval once it reaches the ready state.

Ingest a document as raw text for processing through an automated pipeline. The document progresses through multiple stages (pending → partitioning → indexed → ready) and becomes available for retrieval once it reaches the ready state.

Create a new instruction that automatically extracts structured data from documents as they are created or updated. Instructions apply natural language prompts to documents and store results according to a defined JSON schema.

Generates an OAuth redirect URL for initializing an embedded connector, allowing users to authenticate with third-party services like Google Drive, Notion, Salesforce, and others.

Create a new partition to scope documents, connections, and instructions. Partition names must be lowercase alphanumeric with only underscores and hyphens allowed. Optional resource limits can be defined at creation time.

Generate an LLM-powered response to a query using the deep-search model. Responses can be streamed in real-time or returned synchronously, with optional access to document retrieval tools across specified partitions.

Delete an authenticator connection method. All connections created by this authenticator must be deleted before this operation can succeed.

Schedules a connection for deletion. Optionally preserve associated files (they will be disassociated from the connection) or delete them along with the connection. Deletion is asynchronous and files may remain visible briefly after the request completes.

Permanently delete a document by its unique identifier. Supports both synchronous and asynchronous deletion modes.

Permanently delete an instruction and all entities it generated. This operation cannot be undone.

Permanently deletes a partition and all associated data, including connections, documents, and partition-specific instructions. This operation cannot be undone.

Retrieve a specific connection by its unique identifier. Returns the full connection details including configuration and metadata.

Retrieves aggregated statistics for a specific connection, including total documents, active documents, and total active pages.

Retrieve a specific document by its unique identifier. Returns the full document details including metadata and content.

Retrieves a specific chunk from a document using both the document ID and chunk ID. Use this to fetch individual content segments within a larger document.

Retrieves the content of a specific document chunk in the requested format. Supports streaming media content for audio and video documents, with optional file download capability.

Retrieve the content of a document in your preferred format. Supports multiple media types including JSON (with metadata), plain text, and streaming formats for audio/video content. Non-textual media like images are returned as text descriptions.

Retrieve the original source file of a document. The source varies by origin: uploaded files are returned as-is, URL-sourced documents return the fetched content, and connection-synced documents return the format specific to that connection type (e.g., file from Google Drive, JSON from Salesforce).

Retrieve an LLM-generated summary of a document. The summary is automatically created when the document is first uploaded or updated. Note: This feature is in beta and may change; data files (xls, xlsx, csv, json) and documents exceeding 1M tokens are not supported.

Retrieves a specific element from a document by its unique identifier. Use this to fetch detailed information about an individual element.

Retrieve detailed information about a specific partition, including its usage metrics (document and page counts) and configured limits.

Retrieve a response by its unique identifier. Returns the response data along with its current status: `in_progress` for ongoing processing, `completed` for finished responses, or `failed` for responses that encountered an error.

Retrieve a specific webhook endpoint by its unique identifier. Use this to fetch configuration and status details for a registered webhook.

Ingest a document from a publicly accessible URL for processing and retrieval. The document progresses through multiple processing stages (pending → partitioning → indexed → ready) before becoming available for retrieval, with optional extraction of images, tables, and media content based on the selected partition strategy.

Retrieve a paginated list of all authenticators sorted by creation date in descending order. Use the cursor parameter to navigate through pages when more results are available.

Retrieve all available connection source types (such as Google Drive, Notion, etc.) along with their metadata to understand what integrations can be configured.

Retrieve all connections sorted by creation date in descending order. Results are paginated with a maximum of 100 items per page; use the cursor parameter to fetch subsequent pages when available.

Retrieve all chunks from a document, sorted by index in ascending order. Results are paginated with a maximum of 100 chunks per page; use the cursor parameter to fetch subsequent pages. Documents created before September 18, 2024 that haven't been updated may have chunks with index -1, sorted by ID instead.

Retrieve paginated elements from a document, sorted by index in ascending order. Results are limited to 100 items per page, with cursor-based pagination for accessing subsequent pages.

Retrieve a paginated list of all documents sorted by creation date (newest first). Use the page_size parameter to control results per page and the filter parameter to search by metadata. When more results are available, a cursor will be provided for fetching the next page.

Retrieve all entities extracted from a specific document. Returns a paginated list of entities identified during document processing.

Retrieve all entities that were extracted from a specific instruction. Results are paginated to allow efficient browsing of large entity sets.

Retrieve entity extraction logs for a specific instruction, showing attempt-level results with both successful and unsuccessful outcomes. Results are sorted by creation date in descending order and paginated, with historical data available only from March 6, 2026 onwards.

Retrieve all instructions available in the system. Use this operation to discover and review the complete set of instructions.

Retrieve all partitions sorted alphabetically in ascending order. Results are paginated with a maximum of 100 items per page; use the cursor parameter to fetch subsequent pages when available.

Retrieve all webhook endpoints sorted by creation date in descending order. Results are paginated with a maximum of 100 items per page, and a cursor is provided when additional endpoints are available.

Search and retrieve relevant document chunks based on a semantic query, with optional filtering, reranking, and recency bias to support accurate LLM-based generation and reduce hallucinations.

Immediately schedules a connector to begin syncing data. This operation queues the sync to run as soon as possible.

Update a connection's metadata or partition strategy. Changes take effect after the next sync operation.

Enable or disable a connection to control whether it syncs data. Disabled connections will not perform synchronization operations.

Set or remove page synchronization limits for a connection. When a limit is set, the connection automatically disables after syncing the specified number of pages, though some in-process documents may continue processing.

Replace the file content of an existing document. The uploaded file will be extracted, processed, and indexed for retrieval. Supports text formats (plain text, markdown, email, HTML, XML, JSON, RST, RTF), images (PNG, WebP, JPEG, TIFF, BMP, HEIC), and documents (PDF, Word, Excel, PowerPoint, CSV, EPUB, ODT).

Update a document by ingesting content from a publicly accessible URL. The document progresses through multiple processing states (pending → indexed → ready) before becoming available for retrieval, with optional high-resolution processing for extracting images and tables.

Partially update a document's metadata with new or modified key-value pairs. Reserved keys (document_id, document_type, document_source, document_name, document_uploaded_at) cannot be modified. For connection-managed documents, updates create a metadata overlay applied on each sync.

Replace the raw content of an existing document with new text or JSON data. This operation overwrites the entire document content.

Update an instruction's configuration, including its name, active status, scope, prompt, context template, entity schema, and metadata filters. Changes apply to documents created or updated after the patch is applied.

Update a partition's configuration, including its description and metadata schema. The metadata schema defines an optional subset of document metadata as JSON Schema, useful for LLM-based filter generation.