An assistant is only as helpful as the knowledge it can draw on. A corpus is where that knowledge lives: a shared library of files, web pages, and text that your assistants search before they answer. This guide walks you through creating a corpus and filling it with your first content, and explains why each choice matters along the way.Why build a corpus?#
Out of the box, a language model knows a lot about the world but nothing about your organization: your return policy, your onboarding checklist, last quarter's product changes. When you connect a corpus to an assistant, the assistant retrieves the most relevant passages from your content and grounds its answer in them. The result is responses that are specific, current, and traceable to a source, instead of generic guesses.Because a corpus is separate from any single assistant, you build it once and reuse it. One "HR Policies" corpus can power an HR help desk assistant, an onboarding assistant, and a manager-coaching assistant at the same time. When a policy changes, you update the corpus in one place and every connected assistant picks up the change.When to create a new corpus#
A new knowledge domain. Keep unrelated content apart, for example "Product Documentation" and "Legal Contracts". Focused corpora give more relevant search results and make access control simpler.
Different audiences. If some content is only for a certain team, put it in its own corpus so you can share it with just that team. See Corpus Permissions. Different indexing needs. A small FAQ may be fine on the cost-effective Basic mode, while a large technical library may justify Exhaustive indexing.
Use cases#
Customer support: Upload product manuals, return policies, and troubleshooting articles, so a support assistant answers customer questions accurately, at any hour.Employee help desk: Add the employee handbook, benefits guides, and IT how-tos, so staff get instant answers instead of waiting on a ticket.Sales enablement: Combine pricing sheets, competitor comparisons, and case studies, so sales reps prepare for calls in minutes.
1. Open the Corpora page#
In the sidebar, click Corpus. The Corpora page lists every corpus you have access to, with its owner and indexing mode. Click Create Corpus at the top right.2. Describe your corpus and click Create#
Enter a Corpus Name and a Corpus Description, then choose an Indexing Mode and click Create. A clear name and description help teammates pick the right corpus when they connect it to an assistant.Basic: The most cost-effective option, good when indexing quality isn't a priority.
Moderate: A balanced choice with efficient indexing at a reasonable cost.
Exhaustive: The best accuracy and relevance, at the highest cost. It also supports video files.
You can also add Corpus Tags and turn on Enable PII Redaction here. With PII Redaction on, only TXT, DOCX, and PDF files can be uploaded. See PII Redaction.Not sure which indexing mode to pick? Start with Basic, and move up if answers miss details in your documents. You can change the mode later in the corpus Settings. For a full comparison, see Assistant Corpus Settings. 3. Open your new corpus#
Your corpus now appears in the Corpora list. Use Search corpora... to find it, then click its name.4. Add a knowledge source#
On the Documents tab, click Add. The Add Knowledge Source window shows every way to bring content into the corpus:| Source | Group | Best for |
|---|
| Files | Import Content | Documents you already have, such as PDFs, Word files, spreadsheets, and presentations |
| Website | Import Content | Public web pages, entered one link per line or uploaded as a CSV or TXT list |
| Text Blob | Import Content | Short content you type or paste, such as an FAQ answer or a video transcript |
| SharePoint | Sync Content | Keeping a corpus in step with SharePoint libraries |
| Google Drive | Sync Content | Keeping a corpus in step with Google Drive content |
| Sitemaps | Sync Content | Indexing a whole website from its sitemap |
| Azure NetApp Files | Sync Content | Indexing content stored on Azure NetApp Files volumes |
Click Files to upload documents.5. Upload your files#
Drag files into the window, or click click to browse, then click Upload. The panel on the right lists the supported formats by category (for example Documents, Spreadsheets, Presentations, Email, and Archives). Files can be up to 10 GB, and structured files up to 10 MB.Upload the documents people actually ask about, and remove outdated versions. A smaller, current corpus gives better answers than a large one full of duplicates.
6. Add text directly#
To add content that isn't in a file, click the back arrow, choose Text Blob, type or paste your text, and click Upload Text. Text blobs are ideal for quick facts, policy snippets, or transcripts.Video links can't be indexed. To use a video's content, paste its transcript into a Text Blob instead.
7. Add web pages#
To index web pages, go back and choose Website. Paste one link per line under Enter URLs, or upload a CSV or TXT file with one URL per line under Bulk Upload via File, then click Process & Sync.8. Check that your content is ready#
Close the window to return to the Documents tab. Each item shows its type and status: it starts as Extracting and changes to Completed when it's ready for assistants to use. Use Status and the All, Docs, Links, and Text filters to find items.To remove an item, click its More options menu and choose Delete. The same menu also has Add tag and Open in new window.
Next, connect your corpus to an assistant so it can use this knowledge. See Corpus Connections. FAQs#
Do I need to create a corpus for every assistant?#
No. Every assistant gets a default corpus when it's created. Create a separate corpus when you want to share content across assistants or control access to it separately.Why can't I create a corpus?#
Creating corpora needs a Permission Set with the corpora Create toggle turned on. Ask your administrator if you don't have one. See Corpus Permissions.Who can see my new corpus?#
You're its Owner. Other people get access when you share it with them or their team. See Corpus Access Control.