Requesty review: the 5% AI gateway, tested against its own claims

What Requesty is

Requesty is a hosted gateway: one API key, one endpoint, many providers. Structurally that is the same product as OpenRouter, and the company says so, positioning itself in its own funding announcement as "the premier OpenRouter alternative for EU companies".

Pricing, which is the cleanest in the category

Requesty prices the thing you are buying. There is no top-up fee, no minimum, no seat charge, and no monthly platform bill on the paid tier.

Their own worked example: "a model that costs $10 per 1M tokens from OpenAI costs $10.50 through Requesty." New accounts get $10 in credits at signup.

Features that earn their place

• Routing on three axes: cost, latency, and availability, plus weighted load balancing across models.
• Failover chains. Requesty claims a "99.25% eventual success rate" against 85% for direct API calls. Treat that as a vendor benchmark, not a measurement of your traffic.
• Auto-caching with zero configuration, plus manual cache control.
• BYOK, so you keep negotiated provider rates.
• Observability with cost, latency, time-to-first-token, P50 through P99, error rates, and cache savings, sliced by model, key, user, project, and team.
• Request metadata tagging by feature, user, or trace id for spend attribution.
• Governance: RBAC, approved-model allowlists, PII masking, budget controls, audit trails.
• EU routing. Frankfurt-hosted endpoint at router.eu.requesty.ai.

How many models, actually

Requesty quotes different catalogue sizes depending where you land.

None of these is verifiable from a public models endpoint the way OpenRouter's is, which is a real gap: OpenRouter's /api/v1/models returns a countable list, and that list came back at 415 entries. If catalogue breadth is your deciding factor, ask Requesty for the current list rather than picking a number off a page.

Setup, including Claude Code

Two base URLs, both documented, which is more than most gateways offer. The Anthropic-wire endpoint is the reason Requesty shows up in coding-agent conversations at all.

export OPENAI_BASE_URL="https://router.requesty.ai/v1"
export OPENAI_API_KEY="<your-requesty-key>"

# EU residency
export OPENAI_BASE_URL="https://router.eu.requesty.ai/v1"

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://router.requesty.ai",
    "ANTHROPIC_AUTH_TOKEN": "your_requesty_api_key",
    "ANTHROPIC_MODEL": "anthropic/claude-fable-5"
  }
}

Model ids take the form provider/model-name, and an organization can point a client at a named policy instead with policy/policy-name. Requesty's docs carry one caveat worth repeating verbatim: "We recommend using the ANTHROPIC_MODEL environment variable, and not the /model directive as Claude might not accept it sometimes."

curl -fsSL https://www.requesty.ai/claude/install.sh | bash

Requesty versus OpenRouter

The honest summary: OpenRouter is bigger, more verifiable, and more embedded in the ecosystem. Requesty is cheaper at the small end, better on European data residency, and further along on governance features. If you are an EU company that needs prompts to stay in Frankfurt without an enterprise contract, that single row decides it.

The honest limits

• No public models endpoint. You cannot verify the catalogue the way you can with OpenRouter, and three pages give three numbers.
• Marketing numbers move. Caching savings quoted as 90%, 80%, and 40% on different pages of the same site.
• SOC 2 Type II is in progress, per their own funding post, while the homepage lists SOC2 in a compliance row.
• Smaller than the incumbent. 70,000 developers and a $3M seed against OpenRouter's 250k+ apps. For a component that sits in your request path, vendor durability is a real input.
• Still a third-party processor. EU residency changes where the data sits, not whether an extra party touches it. If that is the blocker, you want a self-hosted gateway.
• Governance is gated. RBAC, PII detection, and guardrails are enterprise-tier, so the pay-as-you-go product is a router, not a control plane.

Questions people ask

Keep reading

Continue through LLM gateways and routing with Helicone review: observability first, gateway second.