General
What exactly does Routor do? Routor sits between your app and every AI provider. You send a request withmodel: "auto", Routor analyzes the prompt in a few milliseconds, ranks candidates by value (quality minus cost), and proxies the request. Your app gets a standard OpenAI-format response back.
How is Routor different from just using GPT-5.4 for everything?
GPT-5.4 is fast and cheap but it fails on complex tasks - detailed code refactors, multi-step reasoning, long-context synthesis. Routor routes simple prompts to an efficient model and complex prompts to a model that can actually handle them. You get cost savings without quality regressions.
Does Routor store my prompts or responses?
Routor does not use your prompts or responses for training. Requests are proxied to providers and not retained after the response is returned. The dashboard shows metadata (model, tier, cost, latency) but not prompt content.
How much latency does routing add?
The routing decision takes a few milliseconds. This is added to your end-to-end latency, but because Routor frequently routes to faster models (Gemini 3.1 Flash-Lite, GLM-4.7 Flash) the net latency is often lower than sending everything to a slow frontier model.
Routing
How does Routor decide which tier to use? Routor uses a 15-dimensional scorer that evaluates: prompt length, vocabulary complexity, presence of code, presence of math/logic, question type, context window needs, and more. The scorer maps the request to one of 5 tiers (NANO through COMPLEX) and ranks candidates in that tier by value (quality + cost savings) while respecting any capability filters you’ve set. Math proofs and formal logic route within COMPLEX to a model built for chain-of-thought work. See How Routor Works for the full breakdown. Can I lock to a specific model? Yes. Create a routing profile with the tier floor and ceiling set to the same tier. If you want to always use Claude Sonnet 4.6 (STANDARD tier), set floor = STANDARD and ceiling = STANDARD. Routor will only pick from STANDARD-tier models. Can I lock to a specific provider? Not directly - Routor routes across providers for reliability. But you can constrain routing to a tier where only one provider’s models currently sit. What if I need vision/tools/audio? Set the required capabilities in your routing profile (Dashboard → Playground → Create Profile). Routor will only route to models that support the capabilities you require. What if the routed model gives a worse answer than I expected? The routing tier represents the minimum quality level needed. If you’re seeing consistently low-quality results, raise the tier floor in your routing profile. You can also test specific prompts in the Playground to see exactly which model gets selected and why.Reliability
What happens if a provider goes down? Routor automatically tries the next best model in the tier pool. If the routed model fails, it retries through a fallback chain until a response succeeds. Transient provider failures are invisible to your app. See Fallback & Auto-Failover for the full behavior. What is the uptime SLA? Routor targets 99.9% uptime. Provider-side failures are handled by the fallback system. Check live provider status on the Provider Health page. Is Routor SOC2 compliant? SOC2 audit is planned. Contact support@routor.io for enterprise security requirements.Billing
What happens when I run out of credits? Requests return a402 Payment Required error. No partial charges, no surprise charges. Top up in Dashboard → Billing to resume.
How do I top up?
Go to Dashboard → Billing → Add Credits. Select a preset amount (₹500, ₹1k, ₹2.5k, ₹5k) or enter a custom INR amount. Payment goes through Razorpay. Credits appear immediately.
Do credits expire?
No. Credits stay in your account indefinitely.
How is the 10% routing margin calculated?
The margin is applied on the model’s cost, not on the credit balance. If a request costs $0.01 in model fees, you are charged $0.011.
See Plans & Pricing for full cost details.
Integration
Do I need to install a new SDK? No. Routor is a drop-in replacement for the OpenAI API. PointbaseURL at https://api.routor.io/v1, swap the key, set model: "auto" - done. See Quickstart.
Does streaming work?
Yes. Routor supports streaming responses (stream: true) exactly as the OpenAI API does. If your deployment has DEBUG_ROUTING=1 set, the X-Routor-* headers are returned with the first chunk.
Does tool calling work?
Yes. Set the tools field in your request as usual. If you create a routing profile, enable Tool calling capability so Routor only routes to models that support it.
Can I use Routor with LangChain / Vercel AI SDK / other frameworks?
Yes. Any framework that supports a custom OpenAI-compatible endpoint works. See:
Profiles
What is a routing profile? A routing profile is a saved set of routing rules - tier range, cost cap, required capabilities, tool-calling quality floor - with a dedicated API key. You create profiles in the Playground and use the profile’s key in your app. How many profiles can I have? Every account includes 1 routing profile - accounts are limited to 1 for now. Can I delete a profile? Yes. Go to Playground → Manage Profiles and click Delete. The profile’s API key is revoked immediately. What happens to requests using a deleted profile’s key? They return401 Unauthorized. Rotate to a new key before deleting the old profile if you have live traffic.