AI Assistant (1.0.0)

Returns version and feature-flag information about the AIAv2 API server.

Analyzes the given documents and suggests redaction annotations based on the redaction criteria. Works with or without a database. Accepts JSON or, for the database-less configuration, multipart/form-data (a file blob plus a data field).

Performs single-shot question-answering across one or more documents, retrieving relevant context to generate an answer. Works with or without a database. Accepts JSON or, for the database-less configuration, multipart/form-data.

Provides an LLM-based summary of what changed plus up to five suggested categories, intended for use as user preferences in a follow-up tagging request. Stateless; no changes are stored.

Categorizes document changes against a list of user-defined labels and returns each change with its matching categories. Stateless; no changes are stored.

OpenAI-compatible chat completions. Supports both simple chat and document-grounded responses. Provide documents to ground the answer on ingested document content.

Stream a run within an existing thread context via Server-Sent Events. State is persisted to the thread, and the run supports interrupts. See the stateless /runs/stream endpoint for the event format.

Execute an agent and stream results via Server-Sent Events. Stateless: no conversation state is persisted. The first event is metadata ({ run_id, attempt }); subsequent events depend on stream_mode (values, messages/partial, messages/complete, updates, debug), ending on completion or an error event. Maximum stream duration is 5 minutes.

Note: JWTs with session_ids claims are not compatible with this endpoint.

Execute an agent to completion and return the final graph state as JSON.

This endpoint runs in stateless mode: an ephemeral thread is created for the execution and deleted after completion. No conversation state is persisted.

Response modes:

• Keep-alive (default): Uses chunked transfer encoding. Newline characters (\n) are sent every second to prevent proxy/load balancer timeouts. The final JSON result is appended after execution completes. The response is transparent to JSON.parse() since leading whitespace is ignored.

• Pure JSON: Send Accept: application/json header. Returns a standard JSON response with no newline padding. Simpler but may timeout on long-running agents behind proxies.

Error handling:

• Pre-execution errors (validation, auth, not found) return standard JSON error responses (400/401/404).
• Execution errors in keep-alive mode return { "__error__": { "error": "...", "message": "..." } } in the response body.
• Execution errors in pure JSON mode return standard JSON error responses.

Headers:

• Content-Location: /runs/{run_id} is set before execution begins.

Maximum execution time is 5 minutes.

Note: JWTs with session_ids claims are not compatible with this endpoint.

Create a new conversation thread.

Search threads with filtering and pagination.

Get thread details.

Delete a thread and its persisted state.

Get the current state of a thread.

Get thread state history with filtering.

List all available assistants with pagination.

Search assistants with advanced filtering.

Get the input, output, state, and context schemas for a specific assistant.

Returns the identifiers of all the documents ingested into the Nutrient AI Assistant database.

If the identifier is listed in the response, the database contains contextual information such document metadata, and document contents, as well as AI synthesized information such as classifications, and summaries.

Deletes the data associated with each layer of the given document, as well as the document default layer (no layer name) itself.

All data including contextual data such as document metadata, and document contents, as well as AI synthesized information such as classifications, and summaries will be deleted.

Deletes the data associated with the given document layer.

All data including contextual data such as document metadata, and document contents, as well as AI synthesized information such as classifications, and summaries will be deleted.

Checks if a document with the given ID and file hash has already been ingested into the Nutrient AI Assistant database.

Ingests a PDF document directly into the Nutrient AI Assistant database. The document will be processed, analyzed, and stored for future retrieval and analysis.

The endpoint expects a raw PDF file in the request body with content-type application/pdf.

Returns permanent and changing IDs for the document.

Ingests a PDF document stored in Nutrient Document Engine into the Nutrient AI Assistant database. The document will be processed, analyzed, and stored for future retrieval and analysis.

Ingests a PDF document stored in Nutrient Document Engine into the Nutrient AI Assistant database. The document will be processed, analyzed, and stored for future retrieval and analysis.

Returns all the session information held in the AI Assistant database.

This includes the session ID, the user ID (if present), and the timestamp of the last message sent.

Deletes all session data for the given Session ID.

This includes all messages and metadata associated with the session. Meaning the chat history will no longer be available.

Returns all the session information for all sessions tagged with a given user ID.

Sessions are tagged with a user ID when the AI Assistant configuration in the Nutrient Web SDK is initialized with a user ID.

Ingests a PDF document without relying on Document Engine. The request body is the raw PDF (application/pdf). Returns the permanent and changing ids and, for client requests, a new JWT scoped to the ingested document.

Ingests a document stored in Nutrient Document Engine. Requires Document Engine.

Ingests a specific layer of a document stored in Nutrient Document Engine. Requires Document Engine.

Checks if a document with the given file hash has already been ingested.

Retrieves the chat history for a specific session.

Responds with the health status of the server.