Website content, knowledge base articles, product catalogs, past conversations, and your own Microsoft AI agents — all retrieved together, ranked by relevance, assembled into one answer.
See it in action How retrieval works →You shouldn't have to choose between your website, your knowledge base, and your product catalog. Velaro's AI bot queries all of them simultaneously — in a single search call with no additional latency.
The web scraper crawls your site and indexes every page. Upload PDFs, policy docs, and manuals. Content stays current on a scheduled refresh.
All plansArticles published in the Velaro Knowledge Base are available to the bot within seconds. The bot draws on exact policy language when it matches the visitor's question.
All plansShopify, BigCommerce, WooCommerce, and NetSuite product data is indexed and kept in sync. The bot can compare specs, look up pricing, and initiate quotes.
Professional+Every resolved conversation that meets a quality bar is automatically analyzed and added to the bot's knowledge. The bot gets smarter from every interaction — no manual effort required.
Professional+Already have a Copilot Studio or Azure AI Foundry agent with access to your internal systems? Velaro forwards questions to it in real time. Your data never leaves your Microsoft tenant.
EnterpriseThe AI can call tools on any MCP-compatible server — CRMs, ERPs, ticketing systems — and include live data in its answer. Connect any system without building a custom integration.
EnterpriseWhen a visitor sends a message, here is what happens — in under 300ms.
The visitor's message is converted to a vector embedding — a numerical representation of its meaning, not just its words.
A single compound query searches your website content, KB articles, product catalog, and conversation learning in parallel. No waterfall, no extra round trips.
If a connected Microsoft AI agent is configured for this workflow step, the question is forwarded in parallel. Its reply is included in the context alongside the indexed results.
The AI reads the combined retrieved context and writes a single, grounded answer. It cites only what was found — no hallucination from thin air.
On Professional and Enterprise plans, Velaro watches every conversation that resolves. The good ones — where the bot or an agent actually helped, with at least four exchanges — are analyzed and their Q&A pairs are added to the bot's knowledge base.
There is no retraining cycle. No manual review step. No exporting data to another system. The knowledge is searchable by the next visitor within minutes of the conversation closing.
Workflows are the sequences of steps your bot follows. Every AI step in a workflow has its own Knowledge Sources panel where you control exactly which sources that step can draw from.
Step 1 (Troubleshoot): searches your KB articles and uploaded support docs only. Step 2 (Escalation check): searches conversation history to see if this customer has reported this before. Two steps, two different knowledge scopes.
Step 1 (Product questions): searches your Shopify catalog and marketing pages. Step 2 (Order status): routes to your Azure AI Foundry agent connected to your ERP. Velaro orchestrates both — the visitor sees one seamless conversation.
If your organization has negotiated Azure pricing or requires data residency for AI inference, Velaro can route all AI calls to your own Azure OpenAI endpoint. You pay Microsoft directly for inference tokens; Velaro manages everything else.
LLM inference — the step where the AI reads your retrieved content and writes the answer. With BYOK on, this call goes to your Azure OpenAI endpoint. You see it in your Azure billing. Velaro's inference cost drops to zero on your account.
Document storage & retrieval — your content lives in Velaro's search infrastructure regardless of BYOK setting. The exception: if you use a connected Microsoft AI Foundry agent, Velaro stores nothing — all retrieval happens in your Azure tenant.
| Source | Where data lives | Latency | Customer data stays in their tenant? | Available on |
|---|---|---|---|---|
| Website scraper | Velaro AI Search | 50–200ms | — | All plans |
| KB articles | Velaro AI Search | 50–200ms | — | All plans |
| Product catalog | Velaro AI Search | 50–200ms | — | Professional+ |
| Conversation Learning | Velaro AI Search | 0ms marginal | — | Professional+ |
| Microsoft AI Agent (Copilot Studio / Azure AI Foundry) | Customer's Azure tenant | 2–15s (live query) | Yes | Enterprise |
| MCP tool server | Customer's server | Varies | Yes | Enterprise |
| BYOK LLM inference | Customer's Azure OpenAI | 1–5s (same as standard) | Yes | Enterprise |
Yes — all enabled sources are searched in a single compound query, not a waterfall. There is no meaningful latency difference between searching one source and searching five. The system uses hybrid retrieval: vector similarity search handles natural language ("what's your return policy") and keyword search handles exact codes and model numbers — both in the same query, merged by relevance score.
Connect your Azure AI Foundry agent or Copilot Studio bot for that workflow step. When the visitor asks an order-status or inventory question, Velaro forwards it to your agent in real time. The agent queries your ERP using its own credentials and returns the answer. Velaro includes that answer in the response. Your ERP data never enters Velaro's systems.
Yes. Conversation Learning is on by default for any step that uses Velaro's knowledge index. You can uncheck it per workflow step in the Knowledge Sources panel. For example, a step that routes to a live Microsoft agent and does not need historical Q&A can have Conversation Learning disabled for that step only. The global subscription flag controls whether the feature is active at all; the per-step setting is a fine-grained opt-out.
No separate configuration. Once your product catalog is indexed (from Shopify, BigCommerce, WooCommerce, or uploaded files), the AI bot can compare products side by side automatically when a visitor asks. The bot retrieves the relevant product entries, compares specs and pricing, and writes a formatted answer. The Quote Maker package adds the ability to initiate a formal quote from within the conversation.
The Bot Playground in Velaro admin shows every document chunk the bot retrieved, labeled by source type and with the original URL or file. You can see exactly which chunk produced the wrong answer, navigate to the source document, and correct it. Conversation Learning entries show the date and channel of the conversation they came from, and can be removed individually if they contain outdated information.
The admin dashboard shows a usage percentage and warns you at 80%. At the hard limit, new scrape jobs are blocked — but existing knowledge keeps working and the bot keeps answering normally. Remove old scrape jobs or outdated files to free space, or contact your account manager to expand your allocation. Storage can be increased on any plan without upgrading your base subscription.
Schedule a demo and we'll walk through your specific knowledge sources and show you what retrieval looks like for your content.
Book a demo