Portkey Alternatives
Six ways to replace <em>Portkey</em> now that it belongs to Palo Alto Networks.
Portkey is now Prisma AIRS AI Gateway, sold by Palo Alto Networks, and its public pricing page now sends you to a demo request. If you want the same gateway on your own servers with no fee, LiteLLM is the replacement. If you care most about latency and agent traffic, Bifrost is open source and adds 11 µs per request on a t3.xlarge instance in its own benchmark. Cloudflare AI Gateway is free and hosted, TrueFoundry fits governed multi-team setups, Kong suits companies already on Kong, and OpenRouter is the simplest switch if you only used the unified API.
The ranking is editorial: a sponsored spot is always flagged on its card, and no affiliate link changes a score.
Why teams look past Portkey
Portkey built one of the most complete AI gateways: a unified API to 1,600+ LLMs, observability, guardrails, governance and prompt management in one platform, with an MIT-licensed open-source gateway underneath. Many teams are happy with it. In 2026 Palo Alto Networks bought it, and that changed what a small or mid-size team gets. These are the reasons we found, each with the fact behind it.
It is now an enterprise security product
No public price grid any more
Log-based plans punished traffic spikes
The open-source gateway is only part of the product
Data residency and an extra hop
Vendor lock-in is now a bigger question
LiteLLM
LiteLLM is the gateway most teams move to when they leave Portkey, and all eight comparison guides we read include it. It comes as a Python SDK and as a proxy server: your apps call one OpenAI-format endpoint, and LiteLLM forwards each request to 1,800+ models from 140+ providers, OpenAI, Anthropic, Vertex AI, Bedrock and the rest. The proxy issues virtual keys per team or user, sets budgets and rate limits on each, and tracks spend per key, team and user across providers. It also speaks MCP and A2A, so the same endpoint can front your agent tools. The project is under the MIT licence, which means you can run it, fork it and change it without asking anyone.
How the price reads against Portkey: the open-source gateway costs nothing, and you pay providers directly with your own keys, so there is no fee on tokens and no log quota that decides your plan. What you pay instead is a server and the time to run it. Enterprise adds SSO/SAML, audit logs, an SLA and the vendor's SOC 2 Type 2 and ISO 27001 reports, on a quote, still on your own infrastructure. The trap: in 2026 two LiteLLM releases published to PyPI (1.82.7 and 1.82.8) carried a credential-stealing payload for about 40 minutes before PyPI quarantined them. The official proxy Docker image was not affected, because it pins its dependencies. Run the image, pin versions and check checksums before each upgrade.
- MIT licence, self-hosted
- 1,800+ models from 140+ providers behind one OpenAI-format endpoint
- Virtual keys, budgets and rate limits per team or user
- MCP and A2A on the same gateway
- ✓No fee on tokens and no log quota
- ✓Your keys and your data stay on your infrastructure
- ✓The closest open-source match to Portkey's gateway features
- ✗You host it, patch it and watch it
- ✗A supply-chain incident on PyPI in 2026: pin versions or use the Docker image
- ✗SSO and audit logs need the Enterprise quote
| Criterion | LiteLLM | Portkey |
|---|---|---|
| Hosting | Self-hosted | Hosted SaaS, now Palo Alto Networks |
| Licence | MIT | MIT gateway, closed control plane |
| Price | Free; Enterprise on quote | No public price grid |
Pick LiteLLM if you have an engineer to own it and want zero fees and full control. If nobody on the team will run a server, take Cloudflare AI Gateway or OpenRouter instead.
Bifrost
Bifrost is an open-source AI gateway written in Go by Maxim AI. It puts 23+ providers (OpenAI, Anthropic, AWS Bedrock, Google Vertex and more) behind one OpenAI-compatible API, with automatic fallbacks between providers and models, load balancing across keys, semantic caching, and governance through virtual keys with budgets and rate limits. It also acts as an MCP gateway, so models can reach tools such as file systems, web search and databases through it. Its selling point is overhead: in its published benchmark at a sustained 5,000 requests per second, the gateway adds 11 µs per request on a t3.xlarge instance, and 59 µs on a smaller t3.medium.
How the price reads against Portkey: the open-source edition is free under Apache 2.0 and covers the core gateway, routing, governance and observability. Enterprise adds private networking, identity providers such as Okta and Entra, audit logs, log export and SLAs, on a licence you request, with a 14-day free trial. Unlike a log-metered plan, nothing in the free edition counts your requests. The catch is maturity: Bifrost is younger than LiteLLM, its observability is thinner than Portkey's dashboards unless you pair it with Maxim's own platform, and its benchmark is the vendor's own. Replay a day of your traffic on your own hardware before you trust the 11 µs.
- 11 µs of overhead per request at 5,000 RPS on a t3.xlarge (vendor benchmark)
- Apache 2.0 licence
- 23+ providers, 1000+ models behind one API
- MCP gateway built in
- ✓Very low added latency
- ✓Fallbacks, load balancing and semantic caching in the free edition
- ✓Single Go binary, quick to deploy
- ✗Younger project with a smaller community than LiteLLM
- ✗Audit logs and identity providers sit in Enterprise
- ✗Deep tracing works best with Maxim's paid platform
| Criterion | Bifrost | Portkey |
|---|---|---|
| Language | Go | TypeScript gateway |
| Licence | Apache 2.0 | MIT gateway, closed control plane |
| Free edition | Full gateway, no request cap | No self-serve plan shown |
Pick Bifrost when latency and agent traffic matter more than a mature dashboard. If you want the largest community and the most integrations, LiteLLM is the safer open-source bet.
Cloudflare AI Gateway
Cloudflare AI Gateway sits in front of your provider calls on Cloudflare's network. You change one line, the base URL, and you get dashboard analytics, caching, rate limiting, request retries and model fallback through model aliases or dynamic routes. It works with your own provider keys, or with Unified Billing, where you buy credits from Cloudflare and provider prices pass through with no markup. DLP scanning is free on all plans, and guardrails run Llama Guard on Workers AI, billed as inference.
How the price reads against Portkey: the core features are free with a Cloudflare account. The cost is in the details. Unified Billing adds a 5% fee on credits, so a $100 purchase is charged $105. Log storage depends on when your account created its first gateway: older accounts keep Legacy Logs, with 100,000 logs in total across all gateways on Workers Free and 10,000,000 per gateway on Workers Paid; accounts that created their first gateway after Cloudflare's 2026 switch store logs in Workers Logs, priced and retained like Workers Logs. That 100,000 is a total, not a monthly allowance: a busy app fills it in days. For long retention, export with Logpush, which is billed by volume.
- Core features free: analytics, caching, rate limiting
- Retries and model fallback through aliases or dynamic routes
- DLP scanning free on all plans
- No markup on Unified Billing, 5% fee on credits
- ✓Nothing to host and nothing to pay to start
- ✓Runs on Cloudflare's edge network
- ✓Works with your own provider keys
- ✗Legacy free log storage is 100,000 logs in total
- ✗No prompt library or deep governance like Portkey Enterprise
- ✗Guardrails and Logpush are billed by usage
| Criterion | Cloudflare AI Gateway | Portkey |
|---|---|---|
| Price to start | Free with a Cloudflare account | Demo request to Palo Alto |
| Hosting | Cloudflare edge | Hosted SaaS |
| Credits | 5% fee on Unified Billing | Your own keys |
Pick Cloudflare when you want Portkey's caching and analytics for free and can live with fewer governance controls. If you need per-team budgets and audit logs, look at TrueFoundry or LiteLLM.
TrueFoundry
TrueFoundry sells an AI gateway for LLMs, agents and MCP servers, aimed at platform teams in larger companies. It covers routing, budgets, rate limits, logs and access control, and on Enterprise it can run in your own VPC or on-prem. It is the vendor that writes most loudly about the Portkey acquisition, so read its marketing with that in mind; its pricing page, though, is clear.
How the price reads against Portkey: Developer is $0, free for up to 3 users, with 10K requests and tool calls per user per month and 7-day log retention. Pro is $25 per user per month, with 20K requests and tool calls per user per month, budget limiting and 30-day log retention; each additional 100K requests costs $20. Pro already lifts the user cap and adds SSO and custom roles; Enterprise brings custom retention and audit exports, VPC, on-prem and data residency, on a custom annual commitment. The trap is the per-seat model: a team of 10 on Pro pays $250 a month before any overage, and the included requests scale with seats, not with traffic. Count your users and your requests before you compare it with a flat gateway.
- Free for up to 3 users
- Pro at $25 per user per month
- LLM and MCP gateway in one product
- 30-day log retention on Pro
- ✓Prices public for Developer and Pro
- ✓Budget limiting, SSO and custom roles on Pro
- ✓Can run in your own cloud
- ✗Per-seat pricing grows with headcount
- ✗+$20 per extra 100K requests on Pro
- ✗VPC, on-prem and audit exports only on Enterprise
| Criterion | TrueFoundry | Portkey |
|---|---|---|
| Pricing model | Per user, plus requests | No public grid |
| Retention | 7 days free, 30 days Pro | Plan-dependent |
| MCP | Included | MCP gateway |
Pick TrueFoundry when governance across teams is the reason you bought Portkey. For a small team, its per-seat price makes LiteLLM or Cloudflare cheaper.
Kong AI Gateway
Kong started as an API gateway and added AI on top: a universal LLM API that routes across OpenAI, Anthropic, Gemini, Bedrock, Azure AI, Mistral and others, plus MCP traffic security and analytics. Kong Gateway itself is open source under Apache 2.0, and the AI features come as plugins. If your platform team already runs Kong for REST traffic, adding LLM routing means new plugins, not a new vendor.
How the price reads against Portkey: the managed Konnect Plus plan starts from $25 a month plus usage, billed monthly, and the AI pieces are priced one by one. A serverless AI gateway control plane costs $25 a month, a hybrid one $200 a month. Plus includes 1 million requests a month, then $200 a month per additional 1 million, up to 10 million. Model proxying costs $100 a month per model, with 5 models at most on Plus. That last line is the hidden cost: a team that routes to five models pays $500 a month before a single request is counted, and a sixth model means Enterprise. Enterprise is custom, billed annually. Kong makes sense when it is already your API layer; as a first gateway for a small AI team, its pricing has more moving parts than any other tool here.
- Apache 2.0 core
- LLM and MCP routing as plugins of an API gateway
- Konnect Plus from $25 a month plus usage
- 1 million requests a month included on Plus
- ✓One gateway for APIs, LLMs and MCP
- ✓Mature plugin ecosystem and policies
- ✓Self-managed option
- ✗$100 a month per proxied model on Konnect
- ✗$200 per extra 1 million requests
- ✗Steeper setup than an AI-only gateway
| Criterion | Kong AI Gateway | Portkey |
|---|---|---|
| Origin | API gateway with AI plugins | AI gateway first |
| Licence | Apache 2.0 core | MIT gateway |
| Model fee | $100/mo per model (Konnect) | None per model |
Pick Kong when it already carries your API traffic. Starting from zero, LiteLLM or Bifrost give you the same routing with fewer line items.
OpenRouter
OpenRouter is a hosted router: one OpenAI-compatible API, 600+ models from 80+ providers, routing and fallbacks, and one prepaid balance instead of an account with each lab. Portkey users look at it when what they mostly used was the unified API, not the deep guardrails or the prompt library. It is the easiest switch on this page: change the base URL, the key and the model names.
How the price reads against Portkey: OpenRouter does not charge a subscription; it charges a platform fee on credits, 5.5% on Standard pay-as-you-go and 8% on Business. On $10,000 of model spend a month, that is $550 on Standard. With your own provider keys (BYOK), the first $25,000 of list-price usage a month carries no fee, then 5%. The Free tier gives 25+ free models from 4 providers, for trials. What you lose against Portkey: a prompt library and deep traces are not its job, and its guardrails are lighter: per-workspace budget caps, model and provider allowlists, regex-based prompt-injection detection and PII redaction. Standard also includes spend controls and admin controls. Pair it with an observability tool if you need them. If your LLM calls run from automation workflows, OpenRouter also plugs into n8n through its OpenRouter chat model node, into Make with the OpenRouter module for Make and into Zapier with the OpenRouter integration for Zapier.
- 600+ models from 80+ providers
- One prepaid balance for every lab
- BYOK free up to $25,000 a month of list-price usage
- 25+ free models
- ✓Fastest migration: base URL, key, model names
- ✓No server to run
- ✓Routing and provider fallbacks built in
- ✗5.5% fee on credits grows with spend
- ✗Lighter guardrails than Portkey, no prompt management
- ✗Data goes through a third party you cannot self-host
| Criterion | OpenRouter | Portkey |
|---|---|---|
| Pricing | 5.5% on credits | No public grid |
| Models | 600+ | 1,600+ (claimed) |
| Guardrails | Lighter: regex injection checks, PII redaction | Built in |
Pick OpenRouter when you used Portkey mainly as a unified API. If you relied on its guardrails, logs or governance, a gateway higher in this ranking fits better. See our OpenRouter alternatives if the fee worries you.
6 Portkey alternatives compared
Each score is the average of five criteria scored out of 5: ease of setup, value for money, routing and features, observability, governance and security. Prices were read on each pricing page in 2026. The edge column says what each tool does differently from Portkey since it joined Palo Alto Networks.
| Best for | Edge over Portkey | Free plan | Team size | Visit | ||||
|---|---|---|---|---|---|---|---|---|
| 1 | LiteLLM | Best self-hosted replacement | Self-hosted, MIT, no fee on tokens | 4.3/5 | Free (self-hosted) | ✓ | Platform teams | Visit → |
| 2 | Bifrost | Best for raw speed and MCP traffic | Go gateway, 11 µs overhead, Apache 2.0 | 4.2/5 | Free (self-hosted) | ✓ | High-volume and agent apps | Visit → |
| 3 | Cloudflare AI Gateway | Best free hosted gateway on your own keys | Free caching and analytics on your keys | 4.1/5 | Free | ✓ | Small teams, Cloudflare users | Visit → |
| 4 | TrueFoundry | Best for governance across many teams | Per-team governance, LLM and MCP | 4.0/5 | Free; Pro $25/user/mo | ✓ | Multi-team companies | Visit → |
| 5 | Kong AI Gateway | Best if you already run Kong | AI plugins on an API gateway you may already run | 3.8/5 | Open source; Plus from $25/mo + usage | ✓ | Existing Kong users | Visit → |
| 6 | OpenRouter | Best hosted catalogue with one bill | 600+ models, one prepaid balance | 3.7/5 | 5.5% fee on credits | ✓ | Developers and prototypes | Visit → |
Prices in USD as shown on each pricing page in 2026. Model usage is billed on top by the providers, except where a tool resells credits (OpenRouter, Cloudflare Unified Billing). Helicone, cited by three of the guides we read, is not ranked: it joined Mintlify in 2026 and has run in maintenance mode since: security updates, new models and bug fixes, with migration help for its customers.
Is your tool missing from this page? Email us: hello@hackceleration.com
You're leaving Portkey because... take this
MIT proxy, virtual keys and budgets, your own provider keys.
Latency and agent traffic matter mostBifrostGo gateway, 11 µs overhead in its benchmark, MCP built in.
You want hosted and free, with no serverCloudflare AI GatewayCaching, rate limits, analytics and fallbacks free.
Many teams need governed access and SSOTrueFoundryFree for 3 users, Pro $25 per user per month.
Kong already runs your API trafficKong AI GatewayLLM and MCP as plugins, one policy layer.
You only used the unified APIOpenRouter600+ models, one key, 5.5% on credits.
How to choose a Portkey alternative
An AI gateway (also called an LLM gateway or LLM proxy) is a layer between your application and the model providers. Your code calls one endpoint, usually in the OpenAI format, and the gateway forwards the request to OpenAI, Anthropic, Google, Bedrock or an open-model host. On the way it adds what providers do not coordinate between them: fallbacks when one fails, retries, load balancing across keys, caching, budgets and rate limits per team, logs and traces, and guardrails that check prompts and answers. Portkey did all of this in one hosted platform. The alternatives fall into three families: self-hosted gateways (LiteLLM, Bifrost, Kong), hosted gateways on top of your keys (Cloudflare, TrueFoundry), and hosted routers that also resell model access (OpenRouter). For the wider field see our best AI gateways; for the models behind them, our best LLM APIs and the cheapest LLM APIs.
What to demand from any gateway
Understand the units you pay for
Compare costs on your real numbers
Self-hosting is a security job
Migrate off Portkey in six steps
Observability may become a separate tool
- OpenAI-compatible endpoint
- Fallbacks across providers
- Budgets per key or team
- Log export and retention you can choose
- Self-hosting or region choice
- Public pricing you can budget
Portkey alternatives: your questions answered
Did Palo Alto Networks acquire Portkey?
Yes. Palo Alto Networks announced the deal in 2026 and has since closed it; Portkey now runs as Prisma AIRS AI Gateway. The portkey.ai home page says Portkey is now Prisma AIRS AI Gateway, generally available for all enterprises, and its pricing link leads to a Palo Alto Networks demo request.
Is Portkey still available as a standalone product?
Not as a self-serve product with public prices. The docs and the gateway still carry the Portkey name next to Prisma AIRS AI Gateway, but new buyers go through Palo Alto Networks sales. If you have a contract, ask in writing how its renewal terms and price will work.
Is Portkey open source?
Partly. The Portkey gateway on GitHub is under the MIT licence and can be self-hosted, which gives you routing, fallbacks and retries. The hosted console, dashboards, prompt management and enterprise controls are not part of that repository.
Can I keep self-hosting the Portkey gateway?
Yes, the MIT licence lets you run and fork it indefinitely. The question is maintenance: the repository announces a 2.0 pre-release that merges the enterprise gateway into open source. Pin the version you run and watch the commit history before you bet on new releases.
What data should I export before leaving Portkey?
Export logs, traces and feedback you need for audits or for replaying incidents, before your plan's retention window deletes them. Also copy your configs, prompt templates with their production version, guardrail rules, budgets and rate limits: none of them move automatically to another gateway.
How much latency does an AI gateway add?
From microseconds to tens of milliseconds, depending on the gateway and where it runs. Bifrost reports 11 µs of overhead per request at a sustained 5,000 requests per second on a t3.xlarge instance in its own benchmark, and 59 µs on a t3.medium. A hosted gateway also adds the network hop to its servers: measure p95 latency on your own traffic before you switch.
Is there a Microsoft equivalent of Portkey?
Yes: the AI gateway in Azure API Management. It is a set of capabilities of API Management, not a separate product, and some of them depend on your tier: token limits and quotas per consumer, semantic caching, load balancing with circuit breakers, content safety checks and token metrics, in front of models in Microsoft Foundry or providers such as Amazon Bedrock, plus MCP servers and A2A agents. It fits a company already on Azure; for multi-cloud routing on your own servers, LiteLLM or Bifrost stay simpler.
Which Portkey alternatives can run fully inside my own network?
LiteLLM, Bifrost and Kong Gateway are open source and run on your own servers; TrueFoundry can also be deployed in your own cloud. Cloudflare AI Gateway and OpenRouter are hosted services, so prompts pass through their infrastructure.