OrcaRouter

OrcaRouter is an OpenAI-compatible AI gateway that grades each prompt in under a millisecond, routes it across 200+ models at provider price with zero token markup, and layers on failover, caching, spend logs, guardrails and an agent firewall for production teams.

OrcaRouter is an AI gateway and LLM router that sits between an application and more than 200 models, sending each request to the model best suited to answer it. Every prompt is graded in under a millisecond and matched to frontier or open-source capacity depending on difficulty, so routine work goes to cheaper models and hard reasoning escalates to stronger ones. The service presents a single OpenAI-compatible endpoint: a developer sets base_url to api.orcarouter.ai/v1 and can call model orcarouter/auto, leaving the choice of model to the router. Token prices are the providers' published rates, with no markup on pay-as-you-go, subscriptions or bring-your-own-key usage. The feature set extends well past model selection. Adaptive routing learns the quality and cost trade-off from live workspace traffic, with policies for cheapest-that-clears-the-bar, highest quality or balanced. Automatic failover, prompt caching and load balancing are built in. Beyond routing, OrcaRouter adds governance: data cloaking replaces names, emails and card numbers before they reach a model; prompt logging is off unless enabled; guardrails block PII, secrets, jailbreaks and prompt injection before billing; and an agent firewall grades each tool or MCP call as allowed, reviewed or blocked. Prompt and response bodies are not logged by default, and retained logs can be sealed with hybrid post-quantum encryption. The product targets engineering teams running AI in production: backend developers who want vendor flexibility without rewriting integrations, platform teams that need spend visibility and per-prompt receipts, and security or compliance owners who must demonstrate controls mapped to 32 frameworks. It competes with plain multi-model aggregators and with managed gateways from cloud vendors, positioning itself on zero token markup, privacy defaults and agent-level controls. Set up is deliberately light, with SDK examples for Python, TypeScript, cURL and agent frameworks such as Claude Code and Codex, plus a public leaderboard and model catalogue for comparison shopping. Teams needing the deepest routing control will still evaluate model behaviour directly; OrcaRouter is best understood as the routing and governance layer rather than a model provider itself.