Drop-in OpenAI compatibility
Use your existing OpenAI SDK and change the base URL. Keep your app logic while routing to lower-cost models.
Built for production AI teams
OpenAI-compatible LLM gateway for lower-cost inference
Route requests across open-weight models like Qwen, DeepSeek, Llama, and Mistral with team budgets, usage logs, fallback routing, and regional deployment options.
No training on customer data
Prompt logging can be disabled
US/EU regional deployment
1 URL
Drop-in OpenAI SDK migration path
5+ models
Open-weight routes for cost and quality tiers
0 training
Customer prompts are not used for model training
Team caps
Budgets, API keys, logs, and model-level limits
Why this exists
LLM costs grow faster than product revenue
Most teams start with a single model API. Then usage grows, costs become unpredictable, API keys get shared across projects, and nobody knows which customer, feature, or model is driving the bill.
OpenRelay AI gives teams one OpenAI-compatible gateway to route requests, set limits, track usage, and switch models without rewriting the app.
Reduce inference cost without changing product logic.
Know which API key, project, model, and customer drives spend.
Keep sensitive routes in approved regions with clear controls.
Gateway controls
One API surface for model routing, budgets, and logs
Built for teams shipping AI products, not chatbot hobby projects. Keep provider choice flexible while giving finance and engineering one clean usage picture.
Model routing and fallback
Route simple tasks to cheaper models and reserve stronger models for complex requests. Add fallback when a provider is slow, down, or rate-limited.
Team budgets and API keys
Create project-level API keys, set monthly budgets, cap model usage, and stop one feature from burning the entire budget.
Usage logs without lock-in
Track requests, tokens, latency, cost, model, provider, and API key. Disable prompt and output logging when privacy matters.
Regional deployment options
Use hosted endpoints or deploy private instances in your preferred region. Keep sensitive data out of regions your customers do not approve.
Clear data handling
Customer prompts are not used for model training. China-based routes are opt-in and disclosed before use.
Qwen
Cost-efficient multilingual routes
DeepSeek
Reasoning and coding tiers
Llama
Open-weight baseline
Mistral
EU-friendly options
BYOK
Bring provider keys when needed
Beta pricing
Start small, then add team controls as usage grows
Use a simple subscription plus service credits model for the beta. Credits are used only for OpenRelay AI API consumption and cannot be transferred or withdrawn.
Developer
$19 /mo
For builders testing lower-cost models.
- OpenAI-compatible API
- Personal API keys
- Usage dashboard
- Email support
Team
$99 /mo
For small teams running AI features in production.
- 5 team seats
- Project API keys
- Monthly budgets
- Request logs
- Priority support
Business
$499 /mo
For teams that need controls, reporting, and reliability.
- Unlimited projects
- Audit logs
- Higher rate limits
- BYOK support
Private
Custom
For dedicated infrastructure and data residency.
- Dedicated endpoint
- US/EU/Singapore regions
- Prompt logging controls
- SLA options
Migration path
Switch one route first, then scale the savings
The beta is built around a practical migration review. Pick one endpoint, test cheaper routes, measure latency and quality, and then expand where the numbers work.
1
Share one endpoint
Tell us the task, current model, monthly tokens, and privacy constraints.
2
Map model routes
Test cheaper open-weight models and define fallback rules for quality.
3
Change base URL
Use your existing OpenAI SDK and point one environment at OpenRelay AI.
4
Measure real usage
Track spend, latency, success rate, and model quality before rollout.
FAQ
Clear answers before you route production traffic
Is this a chatbot?
No. OpenRelay AI is an API gateway for teams building AI products and internal tools.
Do I need to rewrite my app?
Usually no. If you already use the OpenAI SDK, change the base URL and model name.
Do you train on customer data?
No. Customer prompts and outputs are not used for model training.
Are requests sent to China?
Not by default. Hosted routes use approved providers and regions. Any route involving a China-based provider must be explicitly enabled.
Can prompt logging be disabled?
Yes. Teams can disable prompt and output logging while keeping metadata such as tokens, cost, latency, and model.
What happens if a model fails?
You can configure fallback routes to another model or provider.
Cut LLM costs without losing control
Join the beta and get migration help for your first OpenAI-compatible route. We are prioritizing teams with real production usage and clear cost visibility problems.