Augment Context

This guide explains how to give your study's interviewer access to extra knowledge — either through live web search or a private document store you curate.

September 26, 2026

Overview

The Augment Context interface lets you choose how (or whether) the interviewer retrieves information beyond its base study configuration. You can also build and manage a document store by uploading files, tuning how they are split into chunks, and editing the citation metadata attached to each chunk.

Accessing Augment Context

Context in the sidebar (/context/).

Context Enhancement Modes

The first step is choosing how the interviewer should access supplementary knowledge.

Available Modes

  • None: The interviewer relies only on the study configuration. No external sources are consulted.
  • Web Search: The interviewer can look up terms on the open web in real time during a conversation.
  • Document Store: The interviewer retrieves relevant passages from documents you have uploaded and curated.

Saving the Mode

  1. Select the desired mode card
  2. Click Save — a green check indicates success

Only one mode can be active at a time. Switching modes takes effect immediately for new conversations.

Note: The Document Store section is only visible when "Document Store" mode is selected.

Document Store

When "Document Store" mode is active you can upload source documents, review the chunks produced, and edit their metadata.

Uploading a Document

Supported file types: PDF, DOCX, DOC, TXT, MD, HTML, HTM, RTF, PPTX, PPT, XLSX, XLS, CSV, EML, MSG, EPUB, ODT, XML, JSON. Maximum file size is 25 MB.

  1. Click File and select a document from your device
  2. Optionally configure chunking settings and default citation fields (see below)
  3. Click Upload & Chunk

The document is parsed, split into chunks, and each chunk is automatically embedded for retrieval. The new chunks appear in the Document Chunks list immediately.

Chunking Settings

Chunking controls how the uploaded document is divided into retrieval units. Open the Chunking Settings accordion before uploading to customise these parameters. Settings apply only to the next upload.

Basic Settings

  • Chunking Strategy: Controls the primary algorithm used to split the document.
  • By similarity (default): Groups consecutive elements that are semantically related.
  • Basic: Splits purely by character count without considering content boundaries.
  • By title: Starts a new chunk at each detected heading or title element.
  • Max characters: Hard upper limit on the character length of any single chunk.
  • New chunk after N chars: Soft threshold — a new chunk is preferred once this many characters have accumulated, even if the hard limit has not been reached.

Advanced Settings

These options give finer control over edge cases in chunking and PDF processing.

  • Combine under N chars: Elements smaller than this size are merged into the preceding chunk rather than left as standalone entries. Useful for avoiding very small chunks from short headings or captions.
  • Overlap (chars): Number of characters repeated at the start of each new chunk from the end of the previous one. Overlap helps retrieval models find context that straddles a chunk boundary.
  • Overlap all: When enabled, the overlap text is drawn from earlier chunks, not just the immediately preceding one. Increases coverage at the cost of some redundancy.
  • Multipage sections: When enabled (default), a logical section is allowed to continue across page breaks rather than being cut at each page boundary.
  • Split PDF by page: Sends PDF pages to the parser in parallel, which is significantly faster for large PDFs. Enabled by default.
  • Allow failed PDF pages: When split-by-page is active, individual pages that fail to parse are skipped instead of aborting the whole upload. Enabled by default.

Chunking Strategy Guidance

Goal Recommended strategy
General prose documents By similarity (default)
Structured documents with clear headings By title
Fixed-size indexing or debugging Basic

For most research documents, leave the strategy as By similarity and only set Max characters if chunks are coming out too large for your retrieval window.

Default Citation Fields

Before uploading you can fill in citation metadata that will be applied to every chunk produced from that document. Fields left blank are either derived automatically or omitted.

Field Description
Title Document title. Auto-derived from the filename if left empty.
Author Author(s) of the source document.
Date Publication or creation date (e.g. 2024-06-15).
Source Journal, publication, or collection name.
Publisher Publisher or organisation name.
URL Canonical URL of the original source.

Page numbers are populated automatically from the document's structure and do not need to be entered here.

Document Chunks

After uploading, the Document Chunks card lists every chunk stored for the study. Each chunk shows its text content and citation fields, and can be edited or deleted individually.

Chunk Actions

  • Save (per chunk): Saves edits to that chunk's content or citation fields. The button activates as soon as a field is modified. If the text content changes, the embedding is regenerated automatically.
  • Delete (per chunk): Permanently removes that chunk from the document store.
  • Add chunk: Creates a blank chunk manually. You must supply content; an embedding is generated on save.

Per-Chunk Citation Fields

Each chunk exposes the following editable fields:

  • Content: The text that will be retrieved and shown to the interviewer.
  • Title, Author, Date, Source, Publisher: Standard bibliographic metadata.
  • Pages: Comma-separated page numbers (e.g. 1,2,3) extracted from the source document. Can be edited manually.
  • URL: Direct link to the original source.

Selecting and Bulk-Editing Chunks

  1. Tick the checkbox next to each chunk you want to edit, or use Select all
  2. Click Bulk edit citation — the bulk edit modal opens
  3. For each field you want to change, tick its checkbox and enter the new value
  4. To clear a field across selected chunks, tick its checkbox and leave the input empty
  5. Click Apply to selected

Only fields with their checkbox ticked are modified; unticked fields are left as-is on every selected chunk.

Deleting Multiple Chunks

  1. Select the chunks to remove
  2. Click Delete selected

Best Practices

Choosing a Mode

  • Use None when the study's prompt is self-contained and external references could bias responses.
  • Use Web Search for studies about current events or topics where up-to-date information matters.
  • Use Document Store when you need the interviewer to stay grounded in a specific, controlled set of sources (e.g. policy documents, research papers, product manuals).

Preparing Documents

  • Split very large documents into topically coherent files before uploading so chunks stay meaningful.
  • Provide accurate citation metadata upfront — it is easier to fill in the default fields once than to bulk-edit later.
  • Use the By title strategy for documents that have clear section headings; the resulting chunks are easier to review and curate.

Managing the Document Store

  • Review chunks after uploading to remove duplicates, tables of contents, and other low-value content.
  • Use bulk citation editing to standardise author or date fields across an entire uploaded document.
  • Manually add chunks for short passages or definitions that the parser may have merged or discarded.

Troubleshooting

Upload Fails

  • File type not allowed: Convert the document to one of the supported formats (PDF, DOCX, TXT, etc.).
  • File exceeds 25 MB: Split the document into smaller files before uploading.
  • No content was extracted: The file may be scanned or image-only. Try a version with a text layer, or add chunks manually.
  • Parser error: Ensure UNSTRUCTURED_API_KEY is configured in the server environment.

Embeddings Not Generated

  • Confirm that VOYAGE_API_KEY is set in the server environment.
  • Chunks without embeddings will not be retrievable — re-save the chunk to trigger re-embedding.

Chunks Not Appearing After Upload

  • Refresh the page and check that "Document Store" mode is still selected.
  • Check the upload status message below the upload button for error details.

Mode Not Saving

  • Ensure you click Save after selecting a mode card.
  • Check for network errors in the browser console.

Privacy and Security

  • Document store content is scoped to your study and is not shared with other studies or users.
  • Access to the Augment Context page is restricted to authorized users of the study.
  • Follow your organization's data handling policies when uploading documents containing sensitive information.

With MCP

There are no context tools — modes, uploads and chunk editing are dashboard-only. An Interview Flow offers a different way to bring in outside information: a Bring in data (context_fetch) step calls your own API mid-interview and makes the response available to later steps.

Getting Help

For assistance with context augmentation: - Refer to our Help Center - Contact our support team - Check our knowledge base for common scenarios

Last updated: September 2026

Browse Categories

Popular Tags

Recent Articles