Alternatives to Portkey

Portkey alternatives, and whether the acquisition is a reason to move

Palo Alto Networks has completed its acquisition of Portkey, which now ships as Prisma AIRS AI Gateway. That is a real reason to re-evaluate and a poor reason to migrate on its own. This page covers the credible alternatives, and is honest that the acquisition improved Portkey's continuity story as much as it complicated its roadmap.

Reviewed 31 August 2026. Licences read from each project's LICENSE file, stars from the GitHub API, prices from each vendor's own pricing page. Portkey's packaging may have changed since the acquisition closed; verify current terms directly.

Why teams look for an alternative

The acquisition changed the ownership question
Palo Alto Networks has completed its acquisition of Portkey, which is now sold as Prisma AIRS AI Gateway. Acquisitions of infrastructure vendors routinely bring repricing, enterprise repackaging and roadmap reprioritisation. It also brings stronger continuity, so this cuts both ways depending on your buyer.
Self-hosting is positioned on enterprise plans
Portkey is primarily cloud-hosted. If a deployment or data residency constraint is why you are shopping, that is the constraint to test first.
You need scope past the gateway
Agent sessions, isolated code execution, memory and human review sit outside the gateway product.

The alternatives

LiteLLM

The most widely deployed self-hosted AI gateway, with the widest provider coverage anywhere.

  • MIT except the enterprise/ directory, 57,637 GitHub stars (August 2026)
  • Rust core with a Python SDK, describes itself as calling 100+ LLM APIs
  • Enterprise pricing not published
  • Free self-hosting of the core proxy

Best for: Teams whose priority is provider breadth and who can live inside the MIT core.

Watch out: Virtual keys, budgets, RPM and TPM limits, spend by team and SSO are enterprise features.

Full LiteLLM comparison

Bifrost

A Go gateway from Maxim AI with a clean Apache-2.0 licence and a focus on low overhead.

  • Apache-2.0 with no enterprise directory, 7,675 GitHub stars (August 2026)
  • Docker, Kubernetes or a single Go binary
  • OSS tier free forever, Enterprise custom

Best for: Teams who want a fast, simple, cleanly licensed proxy and do not need guardrails or clustering in the free build.

Watch out: Guardrails, Cluster Mode, adaptive load balancing, RBAC, audit logs and SSO are Enterprise.

Full Bifrost comparison

Kong AI Gateway

Enterprise-grade API gateway with AI plugins, and the obvious answer if Kong is already in your path.

  • Core is Apache-2.0, 44,059 GitHub stars (August 2026)
  • Six AI plugins free, seventeen enterprise-only
  • Konnect Plus includes five proxied models, then $100 per month each

Best for: Organisations that already run Kong and want AI traffic governed by the same system.

Watch out: Semantic caching, advanced rate limiting and all guardrails are enterprise. Packaging is changing in Gateway 3.18.

Full Kong AI Gateway comparison

Helicone

Apache-2.0 gateway and observability platform with the lowest-friction integration in this set.

  • Apache-2.0, no enterprise directory, 6,116 GitHub stars (August 2026)
  • Maintenance mode since Mintlify's acquisition announced 3 March 2026; security updates, bug fixes and new models still ship
  • Hobby free with 10,000 requests, Pro $79/month, Team $799/month

Best for: Teams replacing Portkey's observability as much as its routing, who want published tier pricing.

Watch out: Both Helicone and Mintlify describe it as running in maintenance mode after the March 2026 acquisition, which constrains new feature work. Supported on-prem is Enterprise; SOC 2 and HIPAA sit on Team at $799 per month.

Full Helicone comparison

Cloudflare AI Gateway

A hosted platform service with analytics, caching, rate limiting and fallback, free on every Cloudflare plan.

  • Not open source, no self-hosting option
  • Available on all plans including free
  • Nothing to deploy, patch or scale

Best for: Teams leaving Portkey for cost reasons rather than deployment reasons, who already run on Cloudflare.

Watch out: If you are leaving Portkey because it is cloud-hosted, this does not solve that.

Full Cloudflare AI Gateway comparison

Everstack

Ours

A self-hosted-first control plane where the gateway ships with the agent runtime, sandboxes and evaluations.

  • Apache-2.0 across the repository, no separately licensed directory
  • Self-hosting is the default at any tier, not an enterprise plan
  • Guardrails and per-tenant limits in the free build
  • Starter $0, Build $20/month, Scale $100/month (August 2026)

Best for: Teams whose reason for leaving is deployment control, and who also need execution or evaluations in the same place.

Watch out: Portkey's operator UX is more mature than ours, and it now has a large vendor behind it. If continuity is your concern, an independent young company is not obviously the answer.

Full Everstack comparison

How to choose

  1. 01

    Is the acquisition actually your problem?

    Be specific. If the concern is repricing, wait and see what happens to your renewal rather than migrating pre-emptively. If it is roadmap direction, that is a real and durable reason. If it is continuity, the acquisition arguably improved it.

  2. 02

    Do you need self-hosting, or did you just want it?

    If prompts traversing a vendor is acceptable, Cloudflare AI Gateway is free on every plan and requires no infrastructure. Self-hosting is a real ongoing cost and worth taking on only for a real constraint.

  3. 03

    Where does the licence line fall on the alternative?

    Every self-hostable option here gates something. LiteLLM gates governance, Bifrost gates guardrails and high availability, Kong gates caching and guardrails, Helicone gates supported on-prem. Check before adopting.

  4. 04

    How much Portkey-specific configuration do you have?

    Portkey config objects and virtual keys have to be re-expressed as routing policy and API keys wherever you go. The endpoint change is trivial; the configuration port is not automatic. Plan a dual-run rather than a cutover.

Open-source AI infrastructure

Ship the AI stack
you can actually operate.

Start with one model call. Add agents, sandboxes, evaluations, memory, and traces without replacing the foundation underneath them.

Start with $5 creditSelf-host Everstack

One-time starter credit per account on Everstack Cloud. Expires 30 days after signup.