Beta
Discord Download

Portkey alternative: Portkey vs AutoRouter

Portkey is an AI gateway that platform teams put between their apps, coding agents and model providers, with budgets, logs, guardrails and fallbacks. AutoRouter is a Mac app that a company rolls out to every developer’s Mac, adding every leading model to the tools they use and picking one for each turn. This page covers how Portkey works, then how the two compare.

AutoRouter A Mac app your team rolls out through MDM. $0/mo on your plans Enterprise Portkey An AI gateway for platform teams. $49/mo for Production
Price $0 on your plans, Pro $50 a month, Enterprise by seat Free Developer tier, Production $49 a month, Enterprise custom
Billed on Provider’s price plus 10% on paid models Logged requests; you pay your providers directly
Runs as A Mac app, macOS 14 or later A hosted gateway, private cloud, or self-hosted
Agents Codex, Claude Code, Cursor, Claude Desktop, T3 and Terminal Claude Code, Codex, Cursor, Cline, Roo Code, Pi and any OpenAI-compatible agent
Rollout MDM installs it, with nothing for anyone to set up Add providers, issue Portkey keys, point each agent at the gateway
Plans and keys Claude, ChatGPT, Grok and Gemini plans, plus included models Your provider API accounts, such as Anthropic or Bedrock
Model choice Picked for each turn, after your rules The model each agent names, with fallback configs
Governance Allowed models, team budgets, SSO, SCIM, audit log Workspaces, budgets, rate limits, guardrails, SSO, SCIM, audit logs
Spend By team, project or ticket Every request, by team or project
  • Rolled out to every Mac, nothing to build
  • Uses the plans your team already pays for
  • A model picked for each turn
  • Models included, with no keys to hand out

Sources checked November 16, 2026: What is Portkey, Portkey for coding agents, Claude Code guide, Codex guide, Portkey pricing and AutoRouter pricing. AutoRouter wrote this page. See every comparison.

How Portkey works

What the platform is, how it connects to coding agents, and what each plan includes.

A gateway between your apps and providers

Portkey gives one API in front of many model providers, with over 250 models in its introduction and 1,600+ in its Codex guide. Requests pass through its gateway, which adds fallbacks, load balancing, retries, caching, guardrails and a log of every request with its cost, tokens and latency. Its docs now carry the name Prisma AIRS AI Gateway beside Portkey (docs).

For coding agents, Portkey calls itself the governance layer for Claude Code, Codex and others. A platform team keeps provider credentials in its Model Catalog, issues scoped Portkey keys, and sets budgets and rate limits per developer, team or workspace.

How setup works

An admin adds each provider to the Model Catalog with its API key, and creates a Portkey API key. For Claude Code, a developer sets ANTHROPIC_BASE_URL, a Portkey key and provider headers in settings.json, plus the model ID for each tier. For Codex, they add a Portkey provider to config.toml. The npx portkey command sets up routing, MCP servers and team skills for both in one step.

Plans

The open-source gateway is free to host yourself. The hosted Developer tier is free with 10,000 logs a month and isn’t meant for production. Production is $49 a month with 100,000 logs, then $9 per 100,000 more. Enterprise has custom pricing, with SSO, private cloud or VPC hosting, and SOC 2 Type 2, ISO 27001, GDPR and HIPAA (pricing).

What it does well

  • It serves apps, backends and agents through one API, so a company can govern all of its AI traffic in one place.
  • Budgets and rate limits per developer, team or workspace can block requests when a limit is hit.
  • Fallbacks can move Claude Code from Anthropic to Bedrock or Vertex when a provider is down.
  • Guardrails, an MCP gateway, audit logs and SSO with SCIM cover security reviews.
  • It runs on any system that can reach the gateway, including Linux machines and CI.
  • You can self-host the open-source gateway or use private cloud hosting.

Where it gets harder

  • It’s a stack to set up: providers, keys, workspaces, configs and each agent’s settings.
  • It runs on provider API accounts. Its coding agent docs don’t describe using Claude or ChatGPT plans.
  • Each agent uses the model it names. Its coding agent docs describe fallbacks and load balancing, with no model picked by how hard a task is.
  • The Claude Code guide warns that wrong model IDs fail, that the base URL must leave out /v1, and to keep Claude Code updated.
  • Developer and Production keep logs for 3 and 30 days, and longer retention needs Enterprise.

How AutoRouter differs

AutoRouter is for a team that wants its developers’ coding tools governed through an app rollout, without a gateway to build.

Rolled out, with nothing to build

Enterprise rolls AutoRouter out to every Mac through Jamf or another MDM, with nothing for anyone to set up and every Mac running the same way. AutoRouter adds itself to Codex, Claude Code and the other supported apps, and shows up in the model picker each developer already uses. There are no provider keys to hand out or endpoints to configure.

Governance in one place

Admins set allowed models, team budgets and zero data retention for the whole organization, see spend by team, project or ticket for any period, and use SSO, SCIM and an audit log. Enterprise includes everything in Pro and is priced by seat. Contact sales@autorouter.com.

Plans first, then a model for each turn

Your developers’ Claude, ChatGPT, Grok and Gemini plans go to the work that needs them, so they last longer, and work in any harness. Your rules go first, such as “Use Claude Opus 5.5 for UI work.” Otherwise AutoRouter works out how hard each turn is and only considers models that clear the quality bar for it. Every turn shows its model, what it cost and why.

What it can’t do

AutoRouter runs only on macOS 14 or later, so it doesn’t cover Linux machines, CI or your own apps’ API traffic. It sometimes misjudges how hard a task is. Requests pass through AutoRouter’s servers on their way to each provider; see privacy for what is kept.

Which one to choose

Choose Portkey if you need one gateway for apps, agents and CI, want to self-host it, or need its compliance reports and guardrails across all AI traffic.

Choose AutoRouter if your developers work on Macs in Codex, Claude Code and similar tools, and you want an MDM rollout that uses their plans and picks models for them, with no stack to build. See Enterprise.

Setup, side by side

The steps each one’s own docs give, as of November 16, 2026.

Portkey

  1. Add providers In Model Catalog, add each provider with its API key and a slug.
  2. Create keys Generate Portkey API keys, scoped by workspace, with budgets and limits.
  3. Connect agents Run npx portkey, or set the base URL, headers and model IDs in each agent’s config.
  4. Add configs Optionally, add fallback, load balancing and caching configs.

AutoRouter

  1. Roll out Push AutoRouter to every Mac through Jamf or another MDM.
  2. Set policy Choose allowed models, team budgets and zero data retention in one place.
  3. Pick AutoRouter Developers choose AutoRouter in their model picker, or with /model in Claude Code.

Questions

What teams ask about Portkey.

What is Portkey?

An AI gateway and platform that sits between apps or coding agents and model providers. It adds one API, fallbacks, caching, guardrails, budgets and request logs, and its docs now carry the name Prisma AIRS AI Gateway too.

How much does Portkey cost?

The open-source gateway is free to self-host, and the hosted Developer tier is free with 10,000 logs a month. Production is $49 a month with 100,000 logs, then $9 per 100,000 more. Enterprise is custom. You pay providers for tokens separately.

Does Portkey work with Claude Code and Codex?

Yes. Its docs cover Claude Code through Anthropic, Bedrock or Vertex, and Codex through a provider in config.toml, set up by hand or with npx portkey. It also lists Cursor, Cline, Roo Code and Pi.

Is Portkey open source?

Its gateway is open source and free to use. The hosted plans add observability, prompt management and governance on top.

Does Portkey use Claude or ChatGPT plans?

Its coding agent docs describe provider API accounts, such as Anthropic, Bedrock and Vertex, and don’t mention plans. AutoRouter uses Claude, ChatGPT, Grok and Gemini plans first.

How is AutoRouter different from Portkey?

Portkey is a gateway a platform team builds around its provider accounts, for apps and agents on any system. AutoRouter is a Mac app rolled out through MDM that uses your team’s plans, includes models, and picks one for each turn.

Can AutoRouter set team budgets?

Yes. Enterprise sets allowed models, team budgets and zero data retention in one place, and shows spend by team, project or ticket for any period.

What does AutoRouter cost?

Routing across your ChatGPT and Claude plans is free. Pro is $50 a month with $20 of usage, and paid models bill at the provider’s price plus 10%. See pricing, or estimate a month with and without AutoRouter in the savings calculator.