Requesty review

Requesty is a hosted AI gateway and SaaS control plane, not a model, SDK or self-hosted server. An application sends model requests to Requesty’s OpenAI-compatible API; Requesty routes them to a provider and model according to configured policies. Its advertised catalog covers 600+ models.

That places Requesty between application code and model providers. The gateway offers routing by policy, latency, geography or traffic share, plus automatic failover when a provider errors or rate limits. It also centralizes monthly spend limits and request-level observability, including model, provider, tokens, cost, latency and finish reason. Governance features include model and region access rules, PII scanning and SSO.

The architectural tradeoff is centralization: teams can manage provider choices and controls in one service rather than implementing them separately. In return, requests and gateway-specific configuration depend on Requesty’s hosted service. The vendor describes regional routing and EU data residency, but teams should verify the exact regions and controls that apply to their deployment.

The integration surface is an OpenAI-compatible API at https://router.requesty.ai/v1. The homepage’s Python example uses the OpenAI Python SDK, sets base_url to that endpoint and passes a Requesty API key. It then calls chat.completions.create with a provider-qualified model name. Requesty says the API works with any SDK and names Python, Node.js and cURL as integration options.

You can change the model in a request and configure routing policies in Requesty’s app. The vendor says failover and load balancing are handled by the gateway, so application code does not need its own retry logic for those routes. Routing options include traffic shares, latency, geography and automatic selection.

The homepage does not name a CLI or list specific Node.js packages. Confirm SDK support, endpoint compatibility and request behavior in the vendor’s docs before standardizing on an integration. Teams should also check how model-specific parameters, streaming and errors behave through the gateway; these details affect compatibility even when the API follows the OpenAI format.

Requesty has three plans. Free costs $0 per month and includes all free models, routing, caching, fallbacks and EU data residency, with a limit of 200 requests per day. It requires no credit card. That makes it suitable for evaluation or small experiments, but the daily cap makes it a poor fit for steady production traffic.

Pay as you go is usage-based and adds a 5% markup on model spend. It includes access to 600+ models, bring-your-own keys, spend limits, budget caps and the MCP Gateway. The pricing page does not list a monthly or annual billing schedule, a separate per-request fee, or a minimum spend. Confirm how the markup applies when using your own provider keys.

Enterprise pricing is custom. It adds SSO, full RBAC, audit logs, teams, groups, spend controls, guardrails, PII detection, dedicated support and custom SLAs. No add-ons are listed. Requesty does not publish gateway rate limits or overage pricing, so ask about both before estimating a production bill.

Requesty does not publish an open-source licence or a self-hosting option. Treat it as a hosted service unless the vendor confirms otherwise; teams that require a deployable gateway should ask directly about source availability and deployment rights. Data-export options are also not published, so confirm how to export logs, audit records and configuration before adopting it.

The OpenAI-compatible API can reduce application migration work: a client already using the OpenAI SDK can point to a different base URL and select another model. That does not make the full setup portable. Routing policies, spend controls, governance rules and observability data are Requesty-specific surfaces, and the vendor does not describe an export path for them. Keep model calls behind an application-owned interface and document gateway configuration if you want a cleaner exit.

Requesty advertises automatic failover when a provider errors or rate limits, plus regional routing and request-level cost, token and latency tracking. It says teams can set monthly hard limits by API key, user, team or organization. The homepage also advertises 99.25% success with fallback policies, but that figure is not a published uptime SLA. Enterprise includes custom SLAs; the terms are not listed publicly.

The vendor describes EU routing through Frankfurt, US routing through Virginia and APAC routing through Singapore. Its Free plan includes EU data residency, while the pricing page lists gateway limits and a 200-request daily cap for that tier. Requesty does not publish general gateway rate limits, so confirm throughput, timeout, retry and incident-handling expectations before moving production traffic.

A Reddit thread about buying credits raises a procurement concern: commenters objected to Stripe and USD billing when looking for a European alternative. A Requesty representative said the company was evaluating payment options and that most payments were transfers with invoicing. That exchange is not evidence about runtime reliability, but payment and currency details may matter to EU buyers.