What Requesty is
Requesty is a hosted LLM gateway run by Requesty Ltd, a London startup founded by Thibault Jaigu and Daniel Trugman. It raised a $3M seed led by 20VC in September 2025 and has shipped steadily since: an MCP gateway on the pay-as-you-go plan, coding-agent integrations, and open gateway-benchmark datasets published under CC BY 4.0 in April 2026. The product is an OpenAI-compatible endpoint at router.requesty.ai/v1, an Anthropic-style base URL, and a separate EU endpoint in Frankfurt (router.eu.requesty.ai/v1). Its public model list returned 665 models across roughly 31 provider prefixes when we fetched it on August 16, 2026. The homepage reports 70,000+ developers and 90+ billion tokens processed daily; those figures are self-reported. There is no self-hosted option, and the gateway source is not public — the GitHub org contains SDKs and CLI tooling under Apache-2.0 and MIT.
Where Requesty is the better choice
- Spend controls. Hard per-key monthly caps with automatic cutoff, budgets per team, user, and key, alerts by email or webhook, and a management API to adjust limits programmatically. Our spend limits are per API key — daily, weekly, monthly, and lifetime caps with automatic cutoff and email alerts — so Requesty's edge here is team- and user-level budgets and webhook alerts, not the caps themselves.
- Simple, slightly cheaper pricing. A flat 5% markup on the provider's model cost, with a worked example on the pricing page: a model costing $10 per million tokens costs $10.50 through Requesty. No seat fees or minimums, a free tier of 200 requests per day on free models, and $10 of sign-up credits.
- EU residency on every plan. The Frankfurt endpoint is included at every tier, with UK and EU GDPR positioning and a DPA on request; the enterprise plan adds US (Virginia) and APAC (Singapore) residency.
- An enterprise governance checklist. SSO (SAML/OIDC), four-level RBAC, audit logs, guardrails with PII detection, approved-model allowlists, and a 99.99% SLA claim on enterprise. We do not publish a comparable checklist today.
The logging default is the main divide
Requesty's own security FAQ states that "prompt and output logging is enabled by default" on self-serve plans; content is retained encrypted in the EU for up to 30 days. Disabling logging per API key is self-serve, but organisation-wide zero data retention is enabled on written request rather than from the dashboard. On free-plan accounts, Requesty's privacy policy permits using content for its own training. The security page hero says content is never stored; the FAQ on the same page is the accurate statement.
TrustedRouter's realtime path has no equivalent default to turn off: ordinary synchronous and streaming inference does not write prompt or output bodies to persistent storage (privacy, security). What we keep is operational metadata: request ids, selected model and provider, token counts, latency, cost, region, and an API-key hash. Batch is a separate opt-in mode that retains enclave-encrypted artifacts for up to 30 days. Privacy is also routable: trustedrouter/zdr restricts to providers with a cited contractual zero-retention posture, trustedrouter/e2e restricts to confidential-compute providers with provider-side end-to-end encryption (currently Tinfoil and Phala), and trustedrouter/eu forces an EU-focused provider order. The pools are listed on /models.
Trust by policy, trust by measurement
Requesty asks you to trust documents. Those documents are in reasonable shape for a seed-stage company — GDPR posture, quarterly penetration testing, encryption in transit and at rest — but SOC 2 Type II and ISO 27001 are both in progress, with Type II expected Q3 2026 per its own security page, and there is no source code to audit.
Our answer is different in kind, and it has its own limits. Every line that touches your prompt is public: the gateway and control plane are source-available under BUSL-1.1 (converting to Apache-2.0 four years after each release), and the official SDKs are Apache-2.0 or MIT. The running build is verifiable: api.trustedrouter.com serves live hardware attestations from GCP Confidential Space, AWS Nitro Enclaves, and Azure Confidential Containers, and a published verifier script checks the serving code against release digests (trust.trustedrouter.com). The limit: attestation covers our gateway, not downstream model providers. Once a prompt leaves for OpenAI or Anthropic, retention there is a policy question for both products, except on our E2E routes, where the provider's own enclave and encryption mechanisms apply.
Fees on the same basis
Both markups apply to the provider's token price, so the numbers compare directly: Requesty adds 5%; we add 5.5% with a $0.01 per million token floor (/pricing). On the markup alone, Requesty is cheaper. Details that cut both ways: our floor means free-model routes are never billed below $0.01/M, video on TrustedRouter is the provider quote plus 20%, and Requesty's 200-requests-per-day free tier has no TrustedRouter equivalent. On BYOK, Requesty supports four providers (OpenAI, Anthropic, Google AI Studio, and xAI; Vertex is not yet supported) and its docs do not state a BYOK fee; we support BYOK across providers, and our pricing page also publishes no separate BYOK fee number.
What we do not claim
- We are young. Our public repos date from late April 2026 and our public benchmark history starts in June 2026. Requesty's seed closed in September 2025; neither company has a long operating track record.
- We hold no published SOC 2 or HIPAA certification. We publish a DPA, BAA, and subprocessor list, and attestation evidence a reviewer can check directly, but on certifications neither product can hand you a current SOC 2 report today.
- Their live catalog is larger: 665 models to our 561 catalog entries (a count that includes meta routes such as auto and zdr), as of August 16, 2026. We reference more upstream providers (about 49 to their 31).
- We do not match Requesty's team- and user-level budgets, its webhook spend alerts, or its enterprise governance checklist of SSO, RBAC tiers, and PII guardrails.