Data sources

Documentation Index

Fetch the complete documentation index at: https://docs.inboxmate.psquared.dev/llms.txt

Use this file to discover all available pages before exploring further.

Your agent’s responses are only as good as the knowledge you provide. InboxMate supports several types of data sources.

​

Knowledge entries

​

Website scraping

Enter a URL and InboxMate will crawl the page and extract its content into your knowledge base.

• Each scrape counts toward your plan’s monthly scrape limit
• Scrapes capture text content — not images, videos, or JavaScript-rendered content
• Re-scrape pages when content changes to keep your agent up to date

Sitemap scan: Use the sitemap scanner to discover all pages on your website at once. Select which pages to import — InboxMate scrapes each one and creates a knowledge entry per page.

​

PDF uploads

Upload documents like product catalogs, policy documents, or technical manuals. The content is extracted and added to your knowledge base.

• Supported format: PDF
• File size limits vary by plan (up to 25 MB per file)
• Each upload counts toward your monthly PDF limit
• Batch upload: Select multiple PDFs at once to create entries in bulk

​

Custom knowledge items

The most precise way to control your agent’s responses. Create focused entries for specific topics.

​

Vector buckets

Knowledge entries are organized into vector buckets — searchable vector stores that your agent queries at runtime using AI-powered semantic search.

​

How vector buckets work

1. When you add content to a bucket, it’s automatically split into small chunks
2. Each chunk is converted into a vector embedding (a numerical representation of meaning)
3. When a user asks a question, the agent searches the bucket by comparing the question’s embedding with stored chunks
4. The most relevant chunks are returned as context for the agent’s response

​

Adding content to buckets

Click + Element hinzufügen (Add element) on any bucket to add content. All the same content types are available:

• Text — free-form text or FAQ entries
• PDF — upload and auto-extract content
• Multiple PDFs — batch upload multiple documents at once
• Website URL — scrape a single page
• FAQs — structured question-answer pairs
• Sitemap scan — discover and import multiple pages from a website
• Link existing entry — connect an existing knowledge entry to the bucket

All content added to a bucket is automatically chunked and indexed. You can monitor the indexing status (pending, processing, completed, failed) on the bucket detail page.

​

Assigning buckets to agents

Agents can be linked to multiple knowledge buckets, combining different sets of knowledge.

• From the bucket list, click the menu (⋮) on any bucket and select Add to Agent to assign it
• Each bucket card shows which agents are using it
• You can also assign buckets from the agent’s Knowledge settings

​

Best practices for vector buckets

• Create separate buckets for different domains (e.g., “Product docs” vs. “Company policies”)
• Keep content fresh — re-scrape URLs when your website changes
• If search results are poor, check that items are properly indexed (status: “Indexed” with chunk count > 0)
• The search uses semantic similarity — short, focused content chunks work better than huge blocks of text