Integrations
Moving from Helicone
Experiential Labs owns Helicone. When we serve its hostnames, your base URLs, keys, SDKs and headers keep working. This page says what that means for you today and what we do not serve.
What you need to do
Nothing. We are not serving Helicone's hostnames yet. Every Helicone hostname still points at Helicone, so your code, keys and dashboards work exactly as before. We move one hostname at a time, and a hostname can be pointed back to Helicone if anything goes wrong.
After the switch: what stays the same
- The same base URLs, the same
sk-helicone-keys and the same SDKs. - On the provider proxies you keep sending your own provider key on each call. We use it for that call only and serve it only from that provider. We never store it.
- The
Helicone-*headers in the table below. - Your organization and keys are imported for you. Imported Helicone organizations keep Helicone's data terms until you accept ours.
| Host | What happens after the switch |
|---|---|
| oai.helicone.ai, anthropic.helicone.ai, gateway.helicone.ai | Served by us. Your provider key still goes in Authorization and Helicone-Auth carries your Helicone key. |
| ai-gateway.helicone.ai | Served by us with your Helicone key as the bearer. |
| openrouter, deepseek, x, mistral, fireworks, together, cerebras, nebius, novita and google (*.helicone.ai) | Served by us. |
| groq, perplexity, deepinfra, hyperbolic, nvidia, llama, predibase, qstash, firecrawl, vercel, llmmapper, Bedrock, and Azure targets | Sent unchanged to Helicone, which still runs them. |
| api.helicone.ai | The requests, sessions, scores, feedback, webhooks, prompt, user and property calls we answer are served by us. The rest goes to Helicone. |
| Any EU host or EU key | Refused with 421. EU traffic stays on Helicone's EU stack. |
What we do not serve
- EU hosts and EU keys. They stay on Helicone's EU stack.
- Read-only Helicone keys. We do not import them, because our keys have no read-only scope. They keep working for reads on Helicone. You can create a full key here if you need our API.
- Datasets, evals, alerts, the vault, hidden properties and dashboard metrics. Helicone still answers these.
- Helicone's user and property overviews read all time. Ours read at most the last 31 days per call.
- A model we do not carry, on a provider proxy host, is refused with 400
model_not_supported_by_provider. - Your Helicone history is not copied to us yet. It moves only after the legal review is complete and you have had notice.
Helicone headers
A header that has no effect here is accepted and removed before the request reaches a provider, and the response names it in X-Explabs-Ignored-Headers. A safety check you asked for is reported in X-Explabs-Safety-Checks.
| Header | What it does |
|---|---|
| Helicone-Property-* | Becomes a request tag, so it filters in Requests. |
| Helicone-User-Id, Helicone-Session-Id, Helicone-Session-Path, Helicone-Session-Name, Helicone-Request-Id | Become the end user, session, step path, session name and a request id alias. |
| Helicone-Omit-Request, Helicone-Omit-Response | Leave that part out of storage, as X-Explabs-Storage does. |
| Helicone-Retry-*, Helicone-Fallbacks, Helicone-RateLimit-Policy, Helicone-Target-Url, Helicone-OpenAI-Api-Base | Work as on Helicone. |
| Helicone-Cache-Enabled, -Cache-Read, -Cache-Save, -Cache-Seed, -Cache-Bucket-Max-Size (at most 20), -Cache-Ignore-Keys, Cache-Control max-age | The response cache works as on Helicone, in a cache of our own. Nothing is cached when storage is off, zero data retention is required, the key is set to Do not store, or the request omits its request or response. |
| Helicone-Posthog-Key, Helicone-Posthog-Host | Send one event per request to your PostHog, with metadata only (never a prompt or a completion). |
| Helicone-Moderations-Enabled | Checks your last user message with OpenAI moderation, using your own OpenAI key, before the call, as on Helicone. A flagged prompt gets 400 PROMPT_FLAGGED_FOR_MODERATION and is not sent or charged. |
| Helicone-LLM-Security-Enabled, Helicone-Prompt-Security-Enabled, Helicone-LLM-Security-Advanced | Checks your last user message with our own prompt-injection model (and, with Advanced, our content-safety model) before the call. A threat gets 400 PROMPT_THREAT_DETECTED. If the check cannot run, the request is refused with 503 rather than sent unchecked. |
| Helicone-Token-Limit-Exception-Handler, Helicone-Model-Override | truncate, middle-out and fallback work as on Helicone, using our catalog's context windows. Any other value is a 400. |
| Helicone-Lytix-Key, Helicone-Lytix-Host, Helicone-Experiment-*, Helicone-Node-Id | No effect. The response lists them in X-Explabs-Ignored-Headers. |
| Helicone-Prompt-Id with a prompt mode of production or testing | That request is sent to Helicone, which still formats stored prompt templates. |
Storing less
Helicone-Omit-Request: true and Helicone-Omit-Response: true keep working. You can also use our own header, X-Explabs-Storage, or a key setting. See Data controls.