Requesty is a hosted gateway from Requesty Ltd that gives you one OpenAI- and Anthropic-compatible API for 600+ models, with fallbacks, caching and spend tracking. It adds 5% to provider token prices. Its Frankfurt endpoint keeps Requesty's own processing in the EU on every plan; pick an EU-hosted model too if the data must stay there.
Other good fits
Apps that need EU data residency at the gateway: use the EU endpoint and approve only EU-region models.
Pointing Claude Code or an OpenAI-based script at many models through one base URL.
+ Pros
Pick it when you want many models behind one key. Requesty speaks both the OpenAI and the Anthropic API formats, so most tools only need a new base URL and key.
Pick it when you need EU processing. The EU endpoint runs in Frankfurt (AWS eu-central-1), and routing, logging, caching and analytics stay there.
Pick it when an app must survive a provider outage. Fallback and load-balancing policies switch models or providers for you, and you call the policy by name.
Pick it when you want to see cost per user or per feature. Usage analytics, cost tracking and request metadata come built in.
− Cons
Pick something else when you want to run the gateway yourself and pay no markup. LiteLLM is open source.
Pick something else when you'd rather have the model provider itself in the EU and skip the gateway. Scaleway and OVHcloud serve open models from France.
Pick something else when colleagues need a chat window, not an API. Langdock is a hosted workspace; Open WebUI is self-hosted.
One caution: the EU endpoint alone covers only Requesty's processing. If you call a global model, the request leaves the EU for inference.
Avoid if
You want to bring your own key for every model. BYOK covers six providers (OpenAI, Anthropic, Google AI Studio, Vertex AI, Mistral, xAI), one key each.
You must host the gateway yourself. Requesty is a hosted service; LiteLLM is the self-hosted option.
Connection
Set REQUESTY_API_KEY, use https://router.requesty.ai/v1, and send an exact supported model identifier. Policies are configured separately and then used as model IDs; they are not automatically implied by a direct model request.
Billing
Prices checked 23 September 2026 (requesty.ai/pricing):
Plan
Price
What you get
Free
$0, no card
free models only, 200 requests a day, EU data residency
Pay as you go
provider price + 5% (a model at $10 per million tokens costs $10.50)
600+ models, bring your own keys, routing policies, spend limits
Enterprise
custom
SSO, role-based access, audit logs, SLAs
Running effort: someone tops up the balance, sets spend limits, keeps the list of approved models current and reviews the analytics now and then. The bill grows with tokens, so long inputs and big models cost the most.
Data handling
This review did not establish a request-retention configuration. Treat gateway processing and the selected provider's data policy as separate checks before sending non-sample material.
First request
Documentation-based suggestion — not run here.
Read first request
Summarise weekly project notes without the data leaving the EU. A 12-person engineering office wants a five-line summary of each project's notes every Friday.
Sign up (the free tier needs no card) and move to pay as you go so paid models are available.
In the Model Library, approve only EU-region models, for example a Vertex AI Gemini model with an @eu suffix or a Bedrock model in eu-central-1.
Send one project's notes with the prompt: "Summarise in five bullet points: decisions, open risks, next steps. Don't add anything that isn't in the notes."
Open the request log. Check that the model shown is the EU one and note the cost. Then repeat for the other projects.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Based on Requesty's pricing page and documentation (quick start, EU routing, BYOK), checked 23 September 2026. We did not run it for this review.
Requesty pricing · Sep 23, 2026 — Free tier (free models, 200 requests a day), 5% pay-as-you-go markup with $10 to $10.50 example, 600+ models, EU data residency on all plans, Enterprise features; Requesty Ltd.
EU routing · Sep 23, 2026 — EU endpoint in Frankfurt (AWS eu-central-1); processing and storage in EU; global models send inference outside the EU.
Bring your own keys · Sep 23, 2026 — Six supported BYOK providers, at most one key per provider.
Quick start · Sep 23, 2026 — OpenAI-compatible base URL and API key setup.
Further review notes & alternatives
Alternatives and trade-offs
OpenRouter wins when you want the widest provider choice and no markup on tokens. It charges 5.5% on credit purchases, and EU routing needs its 8% Business plan.
LiteLLM wins when you want the gateway on your own server with your own keys.
EUrouter wins when you want a gateway built around models on European servers.
Route requests to many language-model providers through one compatible API.
Use analytics, policy controls, caching, and automatic failover in the gateway.
Select EU residency options when Frankfurt routing fits your deployment needs.