Knowledge Index & File Storage Guide

How Velaro organizes your knowledge

Velaro uses Azure AI Search to power bot answers, Agent Assist suggestions, and auto-learned Q&A. Every site gets an index automatically when you connect a data source (scraper, file upload, KB article, or conversation training). You do not need to create or configure an index — it is provisioned the first time data is ingested.

Shared vs. dedicated physical index

Plan	Index type	What it means
Starter / Professional	Shared	Your data lives in a multi-tenant physical index, isolated by your site ID. Fast setup, no cost overhead.
Enterprise	Dedicated	Your own physical Azure AI Search index. Fully isolated, independently scalable, can use a different embedding model.

Shared indexes are segmented so no site can read another site's data — isolation is enforced at query time by a mandatory site_id filter. If you upgrade from Professional to Enterprise, Velaro migrates your data to a dedicated index in the background and reconnects all your resources (scraper, files, KB articles, conversation training) automatically.

Virtual namespaces within your index

Within your index (shared or dedicated), Velaro separates data by source type using an internal index_name field:

Namespace	What it contains
scraper_pages	Web pages scraped from your site; uploaded documents (AI type)
kb_articles	Knowledge base articles you publish in Velaro
conv_training	High-quality Q&A pairs extracted from past conversations
youtube_content	YouTube transcript content (if connected)

The bot queries across all namespaces that are enabled for your subscription, ranking results by relevance. You do not need to manage these namespaces directly.

Named knowledge indexes — one or many?

You can create multiple named knowledge indexes (your plan limit is shown in the top-right of the Knowledge Index page). By default, bots search all indexes simultaneously. If you assign a specific index to a bot, that bot searches only its assigned index.

Use one index for most situations. A single index handles multiple bots, multiple topics, and large content libraries without any extra configuration. All your bots can query it at the same time.

Use a separate index when content from two areas could contaminate each other's answers. The classic case is serving multiple organizations whose documents are similar in structure but must never mix — for example, two insurance companies whose liability policy language overlaps, or a staffing agency managing separate client handbooks. In these cases, a dedicated index per organization guarantees that a question asked by one organization's visitors can only surface that organization's documents.

Separate indexes are not needed to separate support content from sales content, or to give different bots different focus areas — use the bot's Knowledge Sources panel and workflow system for that instead.

Bots cannot query across indexes: if two bots ever need to combine or compare answers, they must share one index.

How bots are routed to indexes

By default, a bot queries all indexes. To pin a bot to a specific index, go to the bot settings → Knowledge tab → select the index from the Knowledge Index dropdown. Once pinned, the bot ignores all other indexes regardless of what is configured at the account level.

File storage: AI files vs. general files

When you upload a file in Velaro, you choose its purpose:

AI files (uploaded as "AI" type)

• Immediately indexed into your knowledge base vector index.
• The bot can retrieve excerpts from these files to answer visitor questions.
• The bot uses the content of the file to generate answers — it does not send the file as an attachment.
• Use for: product manuals, policy documents, FAQ sheets, training materials you want the bot to draw on.
• File size limit: 25 MB per file. Page and file count limits vary by plan (see your plan's Usage page).

General files (uploaded as "General" type)

• Stored in Velaro's file library but NOT indexed for AI retrieval.
• Agents can manually attach or share a link to these files in any conversation.
• The bot does not automatically surface general files — an agent selects and sends them.
• Use for: contracts, price sheets, signed forms, any document that should only be shared intentionally by a human agent.
• File size limit: 20 MB per file.

How Velaro meters knowledge ingestion

Velaro tracks ingestion by pages processed (across uploaded files and web scraper) and total indexed chunks — not by raw file count alone. Your plan includes a monthly page budget and a total storage ceiling. When you approach a limit, the admin dashboard shows your current usage on the Knowledge → Usage tab. If your use case requires significantly more capacity (for example, a large document corpus across thousands of files), contact your account manager to discuss a custom arrangement.

What counts as a page:

• Uploaded files (PDF, DOCX, XLSX, etc.): 1 page = 1 document page as reported by the file's page count
• Web scraper: 1 page = 1 URL crawled
• Both count against the same monthly page budget

Per-file page cap: Your plan limits how many pages can be indexed from a single file. Pages beyond that cap are not indexed — the first N pages are ingested and the rest are skipped. This prevents a single large document from consuming your entire monthly budget in one upload. If you regularly work with very large documents, contact your account manager.

Controlling what each bot searches (Knowledge Sources)

By default, a bot searches all knowledge sources that are enabled on your plan — website content, Knowledge Base articles, product catalogs, conversation learning, and any files you've uploaded. You can narrow this per bot.

Example: Support bot — check only "Your Website & Files" and "Knowledge Base Articles." Uncheck "Shopify Products" so the bot doesn't surface product listings when customers ask support questions.

Example: Sales bot — check only "Shopify Products" and "Your Website & Files." The bot focuses on your catalog and site content rather than support articles.

To configure knowledge sources: go to your bot → Training tab → Knowledge Sources panel. Each source shows whether it's available on your plan. Grayed-out sources require a plan upgrade. Your bot's own training data (anything you add under the Training Data section) is always searched regardless of these settings.

How to decide: AI file vs. general file

Scenario	Use
Product spec sheet you want the bot to quote from	AI file
Contract template an agent sends after a sale	General file
FAQ document with common support answers	AI file
Price list that changes weekly and must be reviewed before sharing	General file
Installation manual the bot should summarize on request	AI file
Signed agreement the customer requested a copy of	General file

A file cannot be both types at once. If you need the bot to answer questions from a document AND agents to send it as a link, upload it twice — once as AI and once as General.

Conversation training (auto-learn)

When conversation training is enabled, Velaro automatically extracts Q&A pairs from resolved conversations that meet a quality bar (minimum 4 messages, agent participated, CSAT ≥ 4 when collected). These pairs are indexed under the conv_training namespace and used to improve bot answers over time.

Conversation training data is:

• Isolated to your site — no cross-tenant sharing.
• Weighted alongside your other knowledge sources at query time.
• Automatically deduplicated — the same conversation is never re-indexed if it has not changed.

This feature requires the Conversation Training subscription add-on.

What happens as you approach your storage limit

Your plan includes a total indexed chunk ceiling. Velaro tracks this on the Knowledge → Usage tab as a percentage bar. When you reach 80% of your limit, a warning appears in the admin dashboard — you still have headroom, but now is a good time to review what's indexed.

What to do when you get a near-limit warning:

• Review the scraper job list (Knowledge → Web Scraper) and remove any jobs for pages that are no longer needed — old event pages, discontinued product lines, or staging/dev URLs that got indexed by mistake.
• Check uploaded AI files (Knowledge → Files) and delete any outdated policy documents or superseded product specs.
• Each removed job or file frees its indexed chunks immediately after the next scheduled cleanup.

What happens if you hit the hard limit:

• New scrape jobs are blocked until you free space. Existing indexed content still works — the bot continues answering normally.
• You receive the error: "Your plan allows N indexed chunks. Contact support to upgrade your storage."
• Contact your Velaro account manager to increase your limit — it can be raised on any plan without a full plan upgrade.

What does NOT count toward your limit:

• Conversation training Q&A pairs (conv_training namespace) — these count separately under your Conversation Training allocation, not against your chunk ceiling.
• Real-time Copilot Studio / Azure AI Agent queries — no data is stored in Velaro for these.

Reindexing and data freshness

Source	How often re-indexed
Scraper (web pages)	Per your plan: Starter monthly, Professional weekly, Enterprise daily
Uploaded AI files	Immediately on upload; no automatic re-scrape
KB articles	Immediately on publish
Conversation training	Within minutes of conversation resolve

If you delete a file or unpublish a KB article, it is removed from the index at the next scheduled scrape or immediately for KB articles.

Verifying where bot answers come from

Velaro's Bot Playground (Settings → Bots → your bot → Test in Playground) shows not just the AI's answer but every document chunk the bot retrieved to produce it — including which source each chunk came from.

Each retrieved chunk shows:

• Source — whether the answer came from your scraped web pages, a Knowledge Base article, a product catalog, conversation training, or a connected live agent
• Score — how closely the chunk matched the question (higher is better)
• URL — the original page or file the chunk came from

This lets you verify that the bot is pulling from the right sources. If an answer is wrong, you can usually trace it directly to a specific chunk and either update the source content or add a Knowledge Override to correct it.

For connected Copilot Studio agents: the Playground shows which agent was called and includes its raw reply before the bot formatted it for the visitor. If the agent is slow to respond, the playground shows whether the timeout was reached.

For conversation training: chunks from past conversations appear with Source: Conversation Learning and the resolved conversation date. This lets you audit what the bot has learned and remove any conversation whose extracted Q&A is no longer accurate.

Product comparison and quoting from your knowledge index

Once your product catalog, pricing sheets, or spec documents are indexed, your bot can use them for common commerce tasks without any additional configuration:

Product comparison — the bot can compare 2–4 products side by side, pulling specs, pricing, and features from whatever you've indexed (uploaded PDFs, scraped catalog pages, Shopify/BigCommerce products). A visitor saying "what's the difference between Model A and Model B?" triggers a formatted comparison table.

Pricing lookup — the bot retrieves pricing details from your indexed documents, including tiered pricing, bundle pricing, and discounts if those details are in your content.

Quote assistance — for connected ecommerce platforms (BigCommerce, Magento), the bot can initiate a quote directly in the platform. For businesses without a platform connection, the bot can summarize available options and hand off to an agent to finalize.

These capabilities are available under the Quote Maker add-on. No custom development is required — index your content and the bot handles the rest. If your pricing is complex (configurable options, volume tiers, negotiated contracts), upload a structured pricing sheet as an AI file and the bot will reference it.