Route free. Scale securely.
One AI gateway for every LLM: an OpenAI-compatible API for 500+ models, plus an MCP Gateway for every tool. LLM observability, AI governance and managed agents are built in.
Start free
Pay-as-you-go
Route every model and every MCP tool through one governed, OpenAI-compatible endpoint, and scale AI features in production.
500+
Models
30+
Providers
Free BYOK requests
4.5%
Fee orq.ai keys
AI Gateway
MCP Gateway
OpenAI & Anthropic compatible API
Smart Router
Budgets and alerts
PII detection and redaction
30-day retention
Trusted by
Every plan includes
The parts you shouldn't have to pay extra for
One gateway, every model & tool
500+ models from 30+ providers behind one OpenAI-compatible API, plus the MCP Gateway for governed tool calls. Fallbacks, retries, caching and load balancing built in.
Observability from request one
Every routed request becomes a trace: cost, tokens, latency and full payloads, with native OpenTelemetry support.
Runtime controls
Budgets and spend limits, guardrail rules and PII redaction enforced on every request; alert or hard-block before the spend happens.
No lock-in
Integrate with all your existing agents and keys without rebuilds, keep the runtime controls and observability on top, and export all your data easily.
What it actually costs
Estimate your Pay-as-you-go bill
Model traffic is billed at provider rates. This estimates the platform side.
| AI Gateway | |
|---|---|
| 2 seats · 10 included | €0 |
| Platform fee 4.5% | €11.25 |
| 2.5 GB processed · 5 GB included, estimated from usage | €0 |
| AI Observability | |
| 100,000 spans included | €0 |
| Managed Agents | |
| 500 agent runs included | €0 |
How is this calculated?
- Each request through the AI Gateway produces 2 spans and is assumed to average 8,000 tokens across input and output.
- Processed data is estimated at 6 bytes per token with AI Observability off and 8 bytes per token with it on, so a request of 8,000 tokens is 48 KB or 64 KB, respectively. Agent, tool and retrieval spans do not repeat that charge; they count at their JSON size, estimated at 2 KB per span.
- AI Observability is optional. When off, spans are 0 for billing. Processed data remains part of AI Gateway, including the 5 GB allowance.
- Each agent run is assumed to produce 6 spans. Managed Agents is optional. When disabled, run inputs stay preserved but billed runs are 0.
Full comparison
Compare plans
DISCOUNT PROGRAMS
Discounts
Flexible access for early-stage, academic, nonprofit, and open-source teams.
Early-stage startups
Flexible access for teams building their first product.
Research / Students
Support for academic research, teaching, and learning.
Non-profits
Accessible plans for mission-driven organizations.
Open-source projects
Discounted access for open, community-built software.
Reach out to support@orq.ai to apply for a discount. We want all startups, educational users, non-profits and open source projects to build with us.
Usage & Billing
What is a span, and what counts as one?
A span is the unit of observability and billing in Orq.ai. Each discrete operation in your AI application creates one: a model call routed through the AI Gateway, a step inside an agent run (a tool call, reasoning step or retrieval), an evaluator or guardrail execution, or an external span you send via the SDK or OpenTelemetry. A trace is the complete end-to-end interaction; spans are its steps. An agent run is one invocation of an agent and typically contains multiple spans. Billing is based on spans.
What’s the difference between AI Gateway seats and AI Studio seats?
An AI Gateway seat is a team member who holds an API key and routes traffic through the Gateway: engineers shipping features, data scientists running jobs, or anyone calling models from their own code and tools. Gateway seats are unlimited on every plan and never billed per person. What you pay for is their usage, metered as spans and processed data, and you can attribute and budget it per person via Identities. An AI Studio seat is a person working in the Orq.ai interface: writing prompts, configuring agents, running experiments and reading dashboards. Pay-as-you-go includes 10 AI Studio seats. Additional Studio seats cost €35 per seat per month. So a developer who only needs an API key costs you nothing per seat. Add a Studio seat when they need the visual workspace.
What happens when I exceed my included usage?
On Pay-as-you-go, nothing breaks: your traffic keeps routing, and usage beyond the included allowance is billed at the metered rates shown in the table as separate line items on your invoice. You stay in control before that happens: set budgets with alerts, or hard caps per workspace, project, user or API key. Until you add billing details, requests are rate-limited (50 per day) rather than billed.
When am I billed, and how does seat proration work?
Billing runs at the end of each monthly cycle, and included allowances reset at the start of the next. The first 10 seats are included. Beyond those, seats adjust automatically: inviting a member mid-cycle charges a prorated amount immediately; removing one frees the seat at the start of the next cycle.
How can I reduce my bill?
Four ways to keep costs under control:
Bring your own keys: no platform fee at any volume, and model billing stays with your provider.
Smart Router: sends simple prompts to cheaper models, up to 50% lower model spend at ~98% of the quality.
LLM caching: repeated requests return from cache, up to ~95% faster.
Budgets and sampling: cap spend per team or project and sample traces to keep observability volume in check.
Gateway & Models
Do you mark up model traffic?
No. Models are billed at provider list prices; Orq.ai adds no markup. With BYOK, billing stays entirely with your provider and routing is free, with no request limit. If you route on Orq credits instead, a 4.5% platform fee applies on top-ups.
Can I use the AI Gateway without the rest of the platform?
Yes. The Gateway works standalone: swap your base URL, keep your OpenAI-compatible code, and you’re routing in minutes. Observability is captured from the first request, and evals, agents and governance are already there when you need them. No migration, ever.
What latency does the Gateway add?
Smart Router classifies each request in under 40 ms before it reaches a model, and cached responses return up to ~95% faster than a live call. Fallbacks and retries only add latency when a provider actually fails.
Can I use my own private or fine-tuned models?
Yes, on every plan. Connect fine-tuned, self-hosted or private models through Azure AI Foundry, Google Vertex AI or LiteLLM; they join the same catalog as the 500+ public models and work with routing, fallbacks and observability like any other model.
Plan & Subscription
Which plan is right for me?
Pay-as-you-go is right for almost everyone. Start free with the included allowances (100k spans, 5 GB data, 500 agent runs), then pay only for what you use. Every feature is unlocked from day one. Enterprise is for organizations that need custom volumes, enterprise security (SSO, RBAC, audit logs), on-prem or VPC deployment, and dedicated support.
Can I change or cancel my subscription?
Yes, upgrade, downgrade or cancel Pay-as-you-go any time, self-serve. Changes take effect at the end of the current billing cycle. Enterprise agreements run on annual terms.
Do you offer discounts for startups, academic institutions, or NGOs?
Yes, we do. Reach out to us at support@orq.ai, and we’ll prepare a tailored proposal for your organization.
Security & Development
Where does my data live, and is Orq.ai compliant?
Orq.ai is EU-built and EU-hosted, SOC 2 Type II certified, GDPR-compliant with a standard DPA, and aligned with the EU AI Act. Zero-data-retention routing can restrict traffic to providers that store nothing. Full details, reports and controls are available in the Orq.ai Trust Center at trust.orq.ai.
Do you train models on my data?
No. Your prompts, outputs and traces are never used to train models, ours or anyone else’s. You control retention with configurable auto-deletion, and can mask inputs and outputs per request.
Can we self-host or deploy in our VPC?
Yes, Enterprise customers can deploy in their own VPC on AWS or Azure via the marketplaces, or fully on-premise with a single Helm chart, including air-gapped environments. Custom contracts and redlines are available on annual Enterprise agreements.