Keep the proxy. Get six products behind it.
Drop-in proxy replacement — change the base URL, keep your SDK code. You get request/response logging, cost tracking, and caching just like Helicone. Plus red-team security (300+ plugins), LLM firewall, evals (200+ scorers), and prompt IDE in the same workspace.
Same proxy, more surface
What you keep — and what you gain
- ✓ Drop-in proxy — change one URL, keep your SDK
- ✓ Full request / response logging
- ✓ Cost tracking per model, per user
- ✓ Response caching (exact + lexical-similarity)
- ✓ Custom tags for segmentation (flat labels via X-EvalGuard-Tags)
- ✓ Rate limiting per key
- ✓ Prompt-injection screening on the proxy
- ✓ Playground with prompt versioning
- ✓ More generous free tier — 50,000 traces/month here vs 10,000 requests/month on Helicone's Hobby plan (checked 2026-08-09)
- + Firewall on input and output, every provider and every route — 439DLP patterns. Helicone's screen is gated to OpenAI chat-completions in code, and reads only the latest user message.
- + 300+ red-team plugins
- + 200+ prebuilt eval scorers (LLM-as-judge, pairwise, rubric) — none to write
- + Automatic prompt optimizer across 90+ providers
- + OTel trace correlation web→worker→LLM
- + Drift detection with alerting on quality regression
- + 50 compliance frameworks + audit log
Drop-in mapping
Proxy URL mapping
Replace the base URL and swap the Helicone auth header for an EvalGuard key. Your SDK code and provider keys stay the same; Helicone properties map to EvalGuard tags.
| Helicone | EvalGuard | Notes |
|---|---|---|
| https://oai.helicone.ai/v1 | https://evalguard.ai/api/v1/gateway/proxy | Same drop-in base-URL pattern; EvalGuard routes to the right provider based on the model name. |
| https://anthropic.helicone.ai/v1 | https://evalguard.ai/api/v1/gateway/proxy | Same endpoint — claude-* model names auto-route to Anthropic. |
| Authorization: Bearer $OPENAI_KEY | Authorization: Bearer $PROVIDER_KEY | Keep sending your provider key — or store it once in BYOK and drop the header. |
| Helicone-Auth: Bearer $HELICONE_KEY | X-EvalGuard-Key: eg_... | Proxy auth header — keys start with eg_, created at /dashboard/settings. |
| Helicone-Property-*: { key: value } | X-EvalGuard-Tags: tag1,tag2 | EvalGuard tags are a flat label list (not key-value); use them to segment traces. |
Helicone claims on this page were verified against their source — github.com/Helicone/helicone @ 67df07b8d (2026-07-21), plus helicone.ai/pricing raw HTML — on 2026-08-10. Capabilities move; check their current docs before you decide.
One-line migration
Backfill your history
Bring your existing Helicone logs along
The proxy swap captures new traffic. To move the request logs you already have, export them from Helicone and import them with the EvalGuard CLI — no request is dropped, and re-running the same import never double-counts (spans use a stable content hash).
Prints an import summary (spans imported / duplicates skipped / parse errors) and writes the neutral-shape spans to spans.json. The same importer also reads Langfuse and Portkey exports — --from langfuse or --from portkey.
One platform, one bill
Log once. Secure, eval, optimize — everywhere.
Observability is the hook. The platform is why you stay.
Start free — 50K traces / month