Compare gateways

Experiential vs Portkey

Portkey (now Prisma AIRS AI Gateway) is a control layer over your own keys billed per request, while Experiential includes model access, failover and data controls.

Summary

Portkey routes requests to providers through keys you add. It adds configs for fallbacks, load balancing and caching, 50+ guardrails, and request logs. Its docs now carry the banner "Portkey is now Prisma AIRS AI Gateway", linking to Palo Alto Networks, and portkey.ai/pricing redirects to a sales form. Experiential needs no provider accounts and no config objects, and failover is automatic.

import os

from portkey_ai import Portkey

client = Portkey(
    api_key=os.environ["PORTKEY_API_KEY"],
    config="pc-fallback-123",
)
client.chat.completions.create(
    model="@openai-prod/gpt-6-luna",
    messages=[{"role": "user", "content": "Hello"}],
)

Side by side

ExperientialPortkey
HostingHostedHosted, hybrid (gateway in your VPC), or open-source gateway
APIsChat, Responses, Messages, Gemini generateContent, embeddings, batchesChat, Responses, Messages (any provider), embeddings, batches
Model access980 listed models on credits, or your keys3,000+ models through your keys
PricingList price on tokens, no markupPer request: Dev free (10K a month, 3-day logs), Pro $49 (100K, then $9 per 100K), Enterprise custom
FailoverOrdered waterfall, automatic; slow lanes demotedFallback, load-balance and conditional configs; retries, circuit breakers
Zero data retentionPro: org switch, per request, verdict per responseNot documented
Spend limitsPer-key caps, budgets, tier rate limitsBudgets and rate limits per key or provider
Content guardrailsStandard pack or your own classifier50+ checks, partner and custom guardrails
Response cacheNoneSimple and semantic (semantic on Pro and up)
SSONot yet availableEnterprise

Where Experiential is better

  • Failover without configs. Each model has an ordered provider waterfall, and slow providers are demoted automatically. See Routing and failover.
  • Model access included. No provider accounts to open. Credits pay list price, and cost is in each reply.
  • Zero-retention and no-training routing, enforced per org or per request and reported per response. See Data controls.
  • Published pricing, with no per-request platform fee. See Plans and pricing.

When Portkey may fit better

You need a response cache, conditional routing on metadata, prompt templates, a gateway inside your own VPC, or the Palo Alto Networks security stack.

Migration notes

WhatPortkeyExperiential
Base URLhttps://api.portkey.ai/v1https://api.experientiallabs.ai/v1
Authx-portkey-api-key plus x-portkey-provider or a virtual keyAuthorization: Bearer YOUR_API_KEY only
Model ids@provider-slug/modelPlain slug from GET /v1/models
x-portkey-configFallback, load-balance, retry, cacheRemove it; routing and retries are automatic. Fallbacks to a different model are not carried over
override_params, default_paramsSet by configPut the values in the request body
x-portkey-metadataTags; _userX-Explabs-Tags header (a JSON object, up to 16 labels); safety_identifier in the body
x-portkey-trace-idTrace idX-Explabs-Session-Id header
Guardrails in configsInput and output checksPer-key content guardrails

Switch with an agent

Paste this into your coding agent. It inventories the project first and asks before anything it cannot map.

POST/v1/chat/completions
curl https://api.portkey.ai/v1/chat/completions \
  -H "x-portkey-api-key: $PORTKEY_API_KEY" \
  -H "x-portkey-provider: @openai-prod" \
  -H 'x-portkey-metadata: {"_user": "u_42", "team": "search"}' \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-6-luna",
       "messages": [{"role": "user", "content": "Hello"}]}'

Sources, checked 2026-11-19

  1. Feature comparison: plans, request limits, log retention, caching per plan, SSO and VPC on Enterprise.
  2. Feature overview: 3,000+ models, routing, 50+ guardrails, budgets, observability.
  3. AI gateway, universal API, Anthropic, batches.
  4. portkey.ai/pricing: redirects to a Palo Alto Networks contact form. Acquisition: Palo Alto Networks 10-K, FY2026.
  5. Experiential: /llms.txt, GET /api/models.