The open web,
structured for your agents.
Gyrence is a retrieval API for teams building real agents on real web data — not a black box, not a closed SaaS.
A retrieval layer for the open web.
Gyrence is a self-hosted-friendly retrieval API built around five primitives: Search, Gyre, Fetch, Extract, and Map. It's designed for engineers who need programmatic access to the open web — to feed RAG pipelines, power research agents, monitor competitive signals, or build vertical data products — without handing their pipeline to a vendor they can't inspect.
We think the closed-SaaS scraper model is wrong for serious agent workloads. You shouldn't have to choose between "trust a third party with your retrieval logic" and "build everything from scratch." Gyrence ships as an OSS-friendly architecture: predictable primitives, a stable /api/v1 contract, transparent credit accounting, and infrastructure you can reason about.
The bet is simple: the open web is the largest knowledge base AI agents have. It deserves a retrieval layer that treats it that way — structured, queryable, and yours to operate.
Crescentic LLC.
Gyrence is built by Crescentic LLC, a small operator-led shop. We build the tools we wanted to exist while running our own data and AI workloads. That means we optimize for the things that actually matter when you're shipping: honest error envelopes, sane defaults, hard timeouts, and primitives that compose.
We're not trying to be everything. We're trying to be the retrieval layer we'd choose ourselves.
What we don't do (yet).
A few honest limits to know about:
- Sites protected by enterprise bot defenses (Akamai Bot Manager, Cloudflare Advanced, PerimeterX, DataDome) — we detect these challenges and surface them as honest errors rather than saving challenge pages as "content." We don't bypass them, and our browser fleet can't either. Most of the open web works; if you have specific protected sites you need to cover, ask us before signing up and we'll tell you honestly whether they're in scope.
- No async job queue yet. Every request is synchronous within a 25-second budget. Large traversals are coming, but today you orchestrate them client-side.
- No residential proxy pool yet. We route through datacenter egress plus a managed browser worker. Good enough for most sites, not yet good enough for the most aggressively gated ones.
We'd rather tell you this up front than ship marketing copy that breaks on contact with your use case.
Building something where Gyrence might fit?
We want to hear about it.
operator@gyrence.com