AI tool

Portkey

Visit website

Compare gateway reliability controls and log retention with model costs kept outside the platform subscription.

Starting price
Production USD 49/month
Free access
10,000 recorded logs for prototyping
Billing unit
Gateway plan plus model costs
Platform
Hosted gateway; optional self-hosting
Verified
27 Oct 2026

Overview

What you provide

  • Send supported model requests through the configured gateway with provider credentials and routing rules.

What you get

  • The gateway forwards model responses and supplies plan-dependent logs, traces and routing controls.

Choosing this tool

Portkey is useful when an application needs routing, fallbacks and a common view of model calls. Developer is explicitly described as unsuitable for production workloads. Production increases retention and governance, but custom security controls and data-residency guarantees remain an Enterprise discussion. Eden AI and OpenRouter offer alternative unified-model access; compare their fee structure and control surface instead of treating all three as model subscriptions.

A decision worked through

The fixed Production subscription totals USD 588 across twelve monthly payments before usage overages and model bills. A request with retries can generate more upstream work than a single successful user action, so compare dashboard usage and provider invoices before forecasting a total operating cost.

Portkey alternatives for ai gateway routing and observability

Compare Portkey with Eden AI, OpenRouter by the actual task, billing unit and operating responsibility.

ProductPrice / unitFree allowanceInput → outputKey limit
PortkeyRead profileOfficial source ↗Production USD 49/monthDeveloper: 10k recorded logsModel calls gain routing and traces.Inference bills are separate.
Eden AIRead profileOfficial source ↗Model prices + 5.5% credit feeNo general free quota establishedOne API dispatches supported AI tasks.Private deployments use custom pricing.
OpenRouterRead profileOfficial source ↗Standard model rates + 5.5%Eligible free models have limitsCompatible messages become model responses.Credit purchase and BYOK terms differ.

Setup & workflow

  1. Route a non-sensitive test request through one provider.
  2. Simulate a provider failure and inspect retries, fallback and resulting charges.
  3. Roll out routing rules with bounded budgets and an explicit retention policy.
Requirements & installation
  • Map provider keys, fallback order and allowed log fields before enabling production routing.

Limits & review

  • The pricing page mixes recorded-log and request terminology in its allowance and overage sections. Do not convert the listed fee into a guaranteed per-user or per-token price; verify metering and retention against the actual workload.

Your part

  • Set retry limits, budget rules and log redaction before routing sensitive production requests.
Sources & verification5

This profile is based on official sources, not a hands-on product test.

Vendor descriptions and demos document advertised features. Editorial guidance is based on these sources.