The problem#
Directories go stale the day they are published. Spreadsheets of contacts, calendars of events and maps of an industry are rebuilt by hand, again and again, by everyone who needs them. General-purpose search finds pages, not answers; language models answer fluently but cannot say where a fact came from or whether it is still true.
The knowledge is out there. What is missing is infrastructure that reads it continuously, structures it the same way every time, and shows its work.
What Nolvin is#
Nolvin is agentic knowledge graph infrastructure. You describe a domain in one sentence. Agents write a brief, find and validate sources, crawl them politely, and extract the people, objects, locations and events in it — each fact with the page and quote it came from. The graph is assessed, cleaned and extended in cycles, then maintained: re-read on a schedule, paying only for what changed.
Everything is exposed through one API: entities and relationships, neighborhoods, search, timelines and maps, webhooks, share links, exports.
Principles#
- Sourced
- No fact without provenance. Values that must be exact — dates, contact details — are kept only when they appear on the page.
- Structured
- Four entity kinds and fifteen relation types, the same in every domain, so graphs are queryable instead of merely searchable.
- Alive
- A graph is a process, not a snapshot: change detection, diffs and expiry keep it current and let stale facts go.
- Judgement by agents, guarantees by code
- Models decide what matters; deterministic code validates, deduplicates, caps and meters. Nothing an agent proposes is applied unchecked.
- A good citizen of the web
- robots.txt, crawl delays and per-host spacing are honored across every worker. About NolvinBot.
- Cost you can see
- Every model call is metered per domain, with spend caps and a usage API.
Questions#
Is Nolvin a web scraper?
Scraping is the input, not the product. Nolvin decides what to crawl, judges relevance against your rules, extracts typed entities and relationships with their sources, deduplicates and corrects the graph, and keeps it current. What you get is a maintained knowledge graph behind an API, not a pile of pages.
How is this different from asking a chatbot?
A chatbot answers from what it remembers and cannot show where a claim came from. A Nolvin graph is built from pages crawled for your domain, re-read on a schedule, and every fact carries the URL and quote it came from — so it can be checked, cited and kept current.
Who is Nolvin for?
Developers and teams who need structured, current knowledge about a slice of the world — people, organizations, places and events — without building and running a crawling and extraction pipeline themselves: product teams, analysts, community and event sites, and builders of AI agents.
Where does Nolvin run?
On Cloudflare Workers, D1, Queues and Vectorize, with Claude models from Anthropic for extraction and judgement.