Notifications
Clear all
Topic starter
20/07/2026 11:59 am
Just found this in the docs. You can set up a Helicone proxy endpoint for your frontend to call, instead of hitting OpenAI directly.
This seems like it should solve the API key exposure problem. But I have practical questions:
* What's the actual request flow? Does all traffic still route through your servers?
* Are there hard limits on concurrent requests through the proxy?
* How do you handle timeouts for streaming responses?
* In a serverless setup (e.g., Vercel edge functions), does the proxy become a bottleneck?
Looking for anyone running this in production, especially for high-volume RevOps apps where response time is critical.
null