Aeion AI vs Portkey — Native Multi-Provider AI Gateway, Zero Markup
Portkey is a hosted gateway that sits in the request path between your app and 250+ AI providers, charging per-request on top of what you pay the underlying provider. Aeion AI ships the same multi-provider routing + cost tracking + observability + fallback chains as a native part of the platform, with zero markup and no third-party meter in the hot path. 57 adapters today across 9 modalities — LLM, image, audio, video, 3D, embeddings, rerank, STT, TTS.
The Hosted-Gateway Tradeoff
Portkey's pitch is real: a unified API across providers, automatic failover, cost tracking, guardrails, semantic caching, and a polished observability UI. For teams that don't want to operate their own gateway, it's a reasonable choice — at the price of (a) per-request platform markup, (b) every AI call routing through Portkey's infrastructure, (c) provider credentials living on Portkey's servers, (d) your prompts + outputs traversing a third party.
What Aeion AI Actually Ships
The real surface, not the marketing-page version.
The 9 Modalities Aeion AI Covers
Most AI gateways cover LLMs only. Aeion's adapter surface spans 9 modality buckets — the full multi-modal stack.
The Per-Call Cost Math
A realistic enterprise AI deployment running 50M tokens/month across mixed providers.
Migration FAQ
Each Portkey "virtual key" maps to one provider configuration in Aeion, using the underlying provider's real credentials. Migration is a credential-move plus an SDK-call swap. Plan 1-2 sprints for parallel-running before cutover.
Yes — register Portkey via the OpenAI-compatible adapter pointing at Portkey's API. Lets you migrate one model family at a time.
Guardrails are on the roadmap as pluggable adapters. Until then, layer your own filter as a pre-request hook on the engine.
The admin UI surfaces per-tenant cost-by-provider, top consumers, and daily/weekly trends. For deeper analytics, the usage log is queryable from any BI tool.
Portkey's count includes every model variant as a separate "provider." Aeion counts at the SDK/API level — one provider = one adapter, regardless of how many model variants it serves. The 57 adapters cover all the major SDKs in the AI ecosystem; the gap is mostly very niche / regional providers (Chinese-market alternatives, small-shop TTS providers, etc.).
No — the AI gateway is a native part of Aeion OS. There's no separate service to stand up, monitor, or patch; it comes up with the rest of your deployment.