Skip to main content
Velor

Platform · Knowledge Base

Every answer grounded in your current docs. Not yesterday's fine-tune.

Velor retrieves the most relevant passages from your KB on every turn. When the answer isn't there, the agent is instructed to say so and escalate rather than guess.

How it works.

  1. 01

    Index

    Your docs, KB articles, help center pages, and changelogs are chunked and embedded into a vector index.

  2. 02

    Retrieve

    Every incoming message triggers a hybrid search (vector similarity plus keyword match) against the index. The top passages are injected into context.

  3. 03

    Answer

    The agent cites retrieved passages in its response. No training-data drift — uploaded content is live for the next conversation, and crawled sources refresh weekly or on demand from the dashboard.

  4. 04

    Escalate at the boundary

    When no retrieved passage answers the question confidently, the agent is instructed to say so and escalate rather than guess, and the eval shows how often it does.

What makes it precise.

Semantic and keyword retrieval together

Every query is embedded and matched against your KB semantically, and combined with a keyword match. 'How do I reset my password?' finds the right article even if it says 'change your credentials' — not just exact phrase matches.

Relevance floor

Semantic matches below a similarity floor are dropped before the agent sees them, so a weak semantic match is not passed along as if it were an answer.

KB boundary detection

The agent knows the difference between 'I found a clear answer' and 'I'm guessing.' Questions that hit the knowledge boundary are escalated with the attempted retrieval shown — not silently hallucinated.

Indexed in minutes

Connect your help center URL or upload docs directly. Re-indexing happens on a schedule or on-demand when you update content.

What gets indexed.

Supported source types

Help center articles (any provider — plain URLs or crawl)
Documentation sites, including ones built from Markdown or MDX (by URL, sitemap or crawl)
PDF and Word (.docx) uploads
CSV and JSON exports
Notion pages
Text or a URL sent to the API

Each source is chunked, embedded, and stored in Velor's vector store. You can add or remove sources at any time — the index rebuilds automatically.

Run it on your own tickets.

Every workspace starts with a 14-day free trial. No credit card. If you want a guided walkthrough against your data, we're happy to run one with you.