Our own product
How Volator is built
Volator (volator.ai) is retrieval-augmented generation (RAG) end to end. It turns what a business knows, its websites, documents and media, into a knowledge base that agents retrieve from, and runs that knowledge as chat, voice and video agents. Every conversation can then be used and stored back into the knowledge base.
The engineering
Ingestion
Whole websites are crawled, each source with its own crawl settings and an optional schedule to keep it current. PDFs and other documents are extracted with their tables, figures and equations intact, and audio and video are transcribed. The heavy extraction runs as separate services on demand, so the app never waits on it.
A knowledge base you can correct
Every source keeps its extraction history. You open what came out, fix the text or the structured fields by hand, or run the extraction again. When an answer is wrong, you correct the source, not the prompt.
Hybrid search
Content is chunked, embedded and indexed. Search combines semantic and keyword matching, so exact part numbers and loose paraphrases both hit. Only what changed is re-indexed, and stale chunks are removed.
Your vector store, not ours
One knowledge base can feed several destinations at once: the vector store or graph database of your choice, or exports as markdown, llms.txt, JSON and CSV. Each destination refreshes on its own schedule.
Chat, voice and video agents
An agent is defined once and given chat, realtime voice or avatar video as capabilities. Its behaviour is a reusable persona plus a scenario rather than one long prompt, and every turn retrieves its context from the knowledge base. An agent goes on a website with one script tag and a domain allowlist.
Swappable model providers
Chat and extraction run through one model gateway, so the provider is a setting per agent: OpenAI, Anthropic, Google or Mistral. Every call is metered for tokens and cost. The realtime voice pipeline runs on OpenAI models for now.
Conversations written back
When a chat, voice or video session ends, its transcript is stored in the knowledge base set on that agent. The next agent can retrieve it, and you can read what people actually keep asking.
Volator is our own product, in active development and not yet sold to clients.
Ingestion, retrieval and agents on your own data are the same parts we build for clients. AI engineering →
Where it runs
Each business picks a data region when it signs up: the European Union, the United States or Australia. The app keeps a separate database per region and sends every request to its business's own.
Let's talk
Want something like this on your own documents and systems? Tell us what it should answer, and for whom.