Portkey

What it is for

Portkey is an AI gateway: one API in front of 1,600+ models with retries, fallbacks, logs, and prompt templates. Use it when the job is route and observe production LLM calls, not when you need a chat app.

The gateway is open source (Portkey-AI/gateway). Hosted Developer is $0 (10k logs / month). Production is $49/month.

What makes it unique: the cover is Portkey’s official Logs table from the docs — timestamp, model, tokens, and Export Logs on the same chrome you debug.

The homepage now also brands the enterprise path as PRISMA AIRS AI Gateway. The public pricing cards we copied still say Portkey.

Key features

• Universal API — call 1,600+ models through one SDK; model string is @provider-slug/model-name (Getting started).
• Fallbacks / load balancing / retries / timeouts — gateway configs on every public plan card (Pricing).
• Logs and traces — recorded requests with metadata and filters; Developer keeps 3 days of logs, Production 30 days (Pricing).
• Prompt management — templates, playground, versioning; Developer is capped at 3 templates, Production is unlimited (Pricing).
• Guardrails — deterministic checks on Developer; LLM / partner guardrails on Production (Pricing).
• Virtual keys — provider keys stay in Portkey; Production adds RBAC and service-account keys (Pricing).

Who it's for

Best fit: platform teams who will put a gateway in front of several providers and pay when logs exceed 10k.

Poor fit: solo builders who only need a request table and a $0 Hobby card (Helicone).

Pricing

Verified 2026-10-22 (Asia/Shanghai) from portkey.ai/pricing. Exceeding Developer’s log cap does not block requests — extra logs are dropped.

• Open source (self-host) — $0 for the gateway. Universal API, retries, routing, basic dashboard. Community support.
• Developer — $0. 10k recorded logs / month. 3-day log / 30-day metrics retention. 3 prompt templates. Simple cache (1-day TTL). Deterministic guardrails.
• Production — $49/month. 100k logs / month, then $9 per additional 100k (up to 3M). 30-day logs / 90-day metrics. Unlimited templates. Alerts, semantic cache, RBAC, production support.
• Enterprise — custom. SOC 2 / ISO 27001 / GDPR / HIPAA, VPC, private tenancy, SSO.

Limitations / costs

• Developer drops logs after 10k. The call still runs; you just stop seeing it.
• $9/100k overage stops at 3M on the public Production card. After that you talk to sales.
• PRISMA AIRS is the new enterprise name. Confirm which checkout you are on.
• Self-host is not the full SaaS card. Semantic cache, alerts, and compliance live on paid / Enterprise.

Alternatives

• Helicone — gateway + observability with a $79 Pro card and a 10k Hobby tier.
• LiteLLM — open proxy you run yourself if you do not want a hosted control plane.
• OpenRouter — one catalog of models if you want a marketplace, not an enterprise gateway.

Official tutorial

Official Portkey channel video: Portkey Product Overview (oEmbed author: Portkey).

Written path: Getting started with AI Gateway plus Pricing.

Website

https://portkey.ai

Repository

https://github.com/Portkey-AI/gateway

Source

Catalog row curated for AI Tools Nav. Cover is Portkey’s official Logs table (docs). Developer $0 / 10k logs; Production $49 + $9/100k. Tutorial: Product Overview on @portkeyai (oEmbed).

Last verified

2026-10-22 (Asia/Shanghai) editorial check.