Compare
codai vs Portkey
What each one is, how it is priced, where data lives, and when Portkey is the better pick.
Last checked:
In short
Portkey is a US gateway and observability layer over your own provider keys, with an MIT open-source gateway and hosted plans priced by logged requests in USD. codai sells one managed model, priced per verified task in EUR.
When Portkey is the better pick
Pick Portkey if you want detailed per-request logs, conditional routing on metadata and your own provider keys, or an Enterprise hybrid setup with the data plane in your VPC.
| Dimension | codai | Portkey | Sources |
|---|---|---|---|
| What it is / deployment | Managed gateway (SaaS). Runs on Cloud Run in GCP europe-west1 (Belgium). One public model name: codai. | “A unified interface for interacting with over 250 AI models”: hosted tiers plus an open-source gateway. Enterprise hybrid: data plane in your VPC, control plane hosted by Portkey; air-gapped is “no longer offered”. | |
| Open source | No. Not open source and not self-hostable. | Yes: the gateway is MIT (“Copyright (c) 2024 Portkey, Inc”). The README says the “core enterprise gateway is merging into open-source with our 2.0 release”. | |
| Wire formats | OpenAI Chat Completions, OpenAI Responses and Anthropic Messages. | Chat Completions; Responses (“fully Open Responses compliant”, stateful features native-only); Anthropic Messages, where on adapter providers thinking, cache_control and mcp_servers are “silently dropped”. | |
| Pricing model | Monthly plans in EUR: Free (5 verified tasks), BYOK €5, Student €40 (33 tasks), Builder €100 (83), Pro €300 (250). Overage €1.20 per verified task; unverified tasks are billed at raw token cost. | Developer free (10k logs a month, 3-day retention, “Not suitable for production workloads”); Production $49 a month (100k logs, +$9 per extra 100k requests, 30-day retention); Enterprise custom. Model costs go to the providers whose keys you add. | |
| Data residency | Gateway and database in the EU (europe-west1, Belgium). Model inference region depends on the upstream deployment. | Hosted gateway runs “on edge workers throughout the world”. Regional data routing (US/EU/India) and VPC deployments are advertised; the Production plan is “Not recommended” where data residency guarantees are required. The trust portal lists US sub-processors. | |
| Routing | codai routes each request to a tier of Anthropic's current Claude models (Opus 5.5, Sonnet 5.5 or Haiku 5.5) and fails over between approved deployments; the x-codai-routed-to header names the model that answered. | On every plan: conditional routing on metadata, params or path; fallbacks; weighted and sticky load balancing; up to 5 retries. Automatic cheapest-provider switching only in hosted and enterprise versions. | |
| Prompt caching | On by default for every request. | Passes Anthropic cache_control through, including the 1-hour TTL. Simple response cache on all plans (60 s to 90 days); semantic cache on select Enterprise plans. | |
| MCP | Remote MCP server (Streamable HTTP) with OAuth 2.1. | MCP Gateway (mcp.portkey.ai/{server}/mcp) with OAuth, API-key or header auth, per-workspace access control, tool-call logging, rate limits and a registry. | |
| Billing currency / invoices | Prices and charges in EUR; invoices issued by the Romanian seller. | Prices shown in USD. VAT invoices and payment methods are not stated on the pages read. | |
| Company HQ | IMS SRL (Interactive Media Solutions S.R.L.), Târgu Jiu, Romania. | Portkey, Inc, San Francisco, CA, United States. |
Where Portkey’s own pages disagree
The site banner says “Portkey is now PRISMA AIRS AI Gateway” and sales links go to paloaltonetworks.com. We could not confirm the ownership change beyond the banner.
The caching docs say semantic cache is “only available on select Enterprise plans”; the pricing page lists “Simple & Semantic Caching” under Production.
Model counts differ: “1,600+ LLMs” on the home page, “250+ LLMs” and “45+ providers” on GitHub, “3000+ models” in the Messages docs, and both “50+” and “200+ providers” on the integrations page.
Sources
- portkey.ai
- docs.codai.ro/docs/gateway
- codai.ro/subprocessors
- portkey.ai/docs/introduction/what-is-portkey
- portkey.ai/pricing
- portkey.ai/docs/self-hosting/hybrid-deployments/architecture
- github.com/Portkey-AI/gateway/blob/main/LICENSE
- github.com/Portkey-AI/gateway
- docs.codai.ro/docs/gateway/wire-formats
- portkey.ai/docs/product/ai-gateway/responses-api
- portkey.ai/docs/product/ai-gateway/messages-api
- codai.ro/pricing
- portkey.ai/docs/product/model-catalog/integrations
- codai.ro/security
- portkey.ai/portkey-ai-info-page
- trust.portkey.ai
- docs.codai.ro/docs/gateway/models
- portkey.ai/docs/product/ai-gateway/conditional-routing
- portkey.ai/docs/product/ai-gateway/fallbacks
- portkey.ai/docs/product/ai-gateway/load-balancing
- portkey.ai/docs/product/ai-gateway/automatic-retries
- docs.codai.ro/docs/gateway/headers
- portkey.ai/docs/integrations/llms/anthropic/prompt-caching
- portkey.ai/docs/product/ai-gateway/cache-simple-and-semantic
- docs.codai.ro/docs/gateway/clients
- portkey.ai/docs/product/mcp-gateway
- codai.ro/imprint
- portkey.ai/terms
Facts from each vendor’s public pages on October 2, 2026; tell us at [email protected] if something changed.