Start from what you already run
Every other page here starts with a product. This one starts with your infrastructure. Pick something you already run and see which of the 20 products slot into it — and, more usefully, which ones say they do and mean something else.
The middle column is the point of this page. A product lands there when the support is real but not the thing the marketing implies: documented by somebody else, a community package with no support obligation, or a Helm chart nobody names. Those are workable answers. They are not the same answer as a yes, and this page refuses to round them up into one.
The OpenAI SDK
18 fit · 2 with a caveat · 0 noMy code already uses the OpenAI SDK. What can I point it at?
Fits18
- OpenRouter Drop-in: change the base URL and key, nothing else.
- Requesty Drop-in: change the base URL and key, nothing else.
- Braintrust Gateway Drop-in: change the base URL and key, nothing else.
- Helicone Drop-in: change the base URL and key, nothing else.
- Orq.ai Router Drop-in once set up — but first-call work is cloud console setup.
- Portkey Drop-in once set up — but first-call work is sdk install.
- Vercel AI Gateway Drop-in once set up — but first-call work is sdk install.
- Bifrost Drop-in once set up — but first-call work is cli or container.
- LiteLLM Drop-in once set up — but first-call work is cli or container.
- LLM Gateway Drop-in: change the base URL and key, nothing else.
- Amazon Bedrock Drop-in once set up — but first-call work is cloud console setup.
- Azure AI Foundry Drop-in once set up — but first-call work is cloud console setup.
- Cloudflare AI Gateway Drop-in once set up — but first-call work is cloud console setup.
- Google Vertex AI Drop-in once set up — but first-call work is cloud console setup.
- TrueFoundry AI Gateway Drop-in once set up — but first-call work is cloud console setup.
- Fireworks AI Drop-in once set up — but first-call work is sdk install.
- Groq Drop-in once set up — but first-call work is sdk install.
- Together AI Drop-in once set up — but first-call work is sdk install.
With a caveat2
- Kong AI Gateway OpenAI-compatible paths are documented, but not as a general drop-in.
- Apache APISIX AI Gateway OpenAI-compatible paths are documented, but not as a general drop-in.
No0
Everything here has some answer.
The Vercel AI SDK
13 fit · 4 with a caveat · 3 noI build with the AI SDK. Which products have a real provider package?
Only “official provider” means a maintained first-party package. The rest are workarounds of varying quality.
Fits13
- OpenRouter @openrouter/ai-sdk-provider
- Requesty @requesty/ai-sdk
- Helicone @helicone/ai-sdk-provider
- Orq.ai Router @orq-ai/vercel-provider
- Portkey @portkey-ai/vercel-provider
- Vercel AI Gateway @ai-sdk/gateway
- LLM Gateway @llmgateway/ai-sdk-provider
- Amazon Bedrock @ai-sdk/amazon-bedrock
- Cloudflare AI Gateway ai-gateway-provider
- Google Vertex AI @ai-sdk/google-vertex
- Fireworks AI @ai-sdk/fireworks
- Groq @ai-sdk/groq
- Together AI @ai-sdk/togetherai
With a caveat4
- Braintrust Gateway Documented integration, not a provider
- Apache APISIX AI Gateway Via the OpenAI provider
- Bifrost Claimed compatible, no package named
- Azure AI Foundry Community provider package
No3
- Kong AI Gateway No AI SDK route documented.
- LiteLLM No AI SDK route documented.
- TrueFoundry AI Gateway No AI SDK route documented.
Cloudflare Workers
4 fit · 2 with a caveat · 14 noI run on Workers. What can I actually call from the edge?
Three products market Cloudflare Workers heavily and mean their own gateway runs there. That does not help you call them from your Worker, so they are not counted as a fit.
Fits4
- Requesty Calling it from a Worker is documented.
- Braintrust Gateway Calling it from a Worker is documented.
- Cloudflare AI Gateway Calling it from a Worker is documented.
- Google Vertex AI Calling it from a Worker is documented.
With a caveat2
- Amazon Bedrock Documented by Cloudflare, not the vendor.
- Groq Documented by Cloudflare, not the vendor.
No14
- OpenRouter Their own gateway runs on Workers. That is not the same as edge support for your code.
- Helicone Their own gateway runs on Workers. That is not the same as edge support for your code.
- Kong AI Gateway No Workers guidance published.
- Orq.ai Router No Workers guidance published.
- Portkey Their own gateway runs on Workers. That is not the same as edge support for your code.
- Vercel AI Gateway No Workers guidance published.
- Apache APISIX AI Gateway No Workers guidance published.
- Bifrost No Workers guidance published.
- LiteLLM Workers AI is an upstream model here, which is the opposite direction.
- LLM Gateway No Workers guidance published.
- Azure AI Foundry No Workers guidance published.
- TrueFoundry AI Gateway No Workers guidance published.
- Fireworks AI No Workers guidance published.
- Together AI No Workers guidance published.
Kubernetes
11 fit · 2 with a caveat · 7 noI have a cluster and want this inside it. What can I install?
Fits11
- Requesty Helm repo https://helm.requesty.ai (chart name not stated)
- Braintrust Gateway braintrust from oci://public.ecr.aws/braintrust/helm/braintrust
- Helicone helicone-core, helicone-infrastructure, helicone-monitoring, helicone-argocd
- Kong AI Gateway kong/ingress and kong/kong from https://charts.konghq.com
- Orq.ai Router single Orq chart, Kubernetes 1.28+
- Portkey charts/portkey-gateway and charts/portkey-app in Portkey-AI/helm
- Apache APISIX AI Gateway apache/apisix-helm-chart; API7 ships api7/gateway from https://charts.api7.ai
- Bifrost bifrost/bifrost from https://maximhq.github.io/bifrost/helm-charts
- LiteLLM litellm-helm at oci://ghcr.io/berriai/litellm-helm — the GitHub page describes it as community maintained
- LLM Gateway OCI chart on GHCR; deploys gateway, API, UI and worker, and needs PostgreSQL plus Redis
- TrueFoundry AI Gateway tfy-llm-gateway for the Gateway Plane, truefoundry for the full control plane plus gateway plane
With a caveat2
- Azure AI Foundry Helm is supported but no chart is named.
- Fireworks AI Helm is supported but no chart is named.
No7
- OpenRouter No Kubernetes deployment published.
- Vercel AI Gateway No Kubernetes deployment published.
- Amazon Bedrock Only generic cluster docs — nothing to install for this product.
- Cloudflare AI Gateway No Kubernetes deployment published.
- Google Vertex AI Only generic cluster docs — nothing to install for this product.
- Groq No Kubernetes deployment published.
- Together AI Only generic cluster docs — nothing to install for this product.
Terraform or OpenTofu
4 fit · 11 with a caveat · 5 noEverything I run is in code. What can I declare?
Fits4
- Kong AI Gateway Kong/kong-gateway, konnect, konnect-beta and kong-mesh
- Orq.ai Router orq-ai/orq 0.1.0
- Portkey Portkey-AI/portkey
- Cloudflare AI Gateway cloudflare/cloudflare with cloudflare_ai_gateway and cloudflare_ai_gateway_dynamic_routing
With a caveat11
- OpenRouter Community provider
- Braintrust Gateway Official modules
- Helicone Terraform code in the repo only
- Bifrost Official modules
- LiteLLM Official modules
- LLM Gateway Official modules
- Amazon Bedrock Resources in the cloud’s provider
- Azure AI Foundry Resources in the cloud’s provider
- Google Vertex AI Resources in the cloud’s provider
- TrueFoundry AI Gateway Terraform code in the repo only
- Fireworks AI Official modules
No5
- Requesty Nothing published for Terraform.
- Vercel AI Gateway Nothing published for Terraform.
- Apache APISIX AI Gateway Nothing published for Terraform.
- Groq Nothing published for Terraform.
- Together AI Nothing published for Terraform.
An existing API gateway
2 fit · 6 with a caveat · 12 noI already run Kong or APISIX. Can AI traffic go through it?
Fits2
- Kong AI Gateway This is that gateway — AI traffic becomes a plugin, not a new hop.
- Apache APISIX AI Gateway This is that gateway — AI traffic becomes a plugin, not a new hop.
With a caveat6
- OpenRouter Documented by Apache APISIX, not this vendor.
- Braintrust Gateway Service mesh routing documented
- Amazon Bedrock The cloud ships its own AI gateway
- Azure AI Foundry Documented by Kong, not this vendor.
- Google Vertex AI The cloud ships its own AI gateway
- Fireworks AI Documented by API7, not this vendor.
No12
- Requesty Nothing published about running behind your gateway.
- Helicone Nothing published about running behind your gateway.
- Orq.ai Router Nothing published about running behind your gateway.
- Portkey Nothing published about running behind your gateway.
- Vercel AI Gateway Nothing published about running behind your gateway.
- Bifrost Nothing published about running behind your gateway.
- LiteLLM Nothing published about running behind your gateway.
- LLM Gateway Nothing published about running behind your gateway.
- Cloudflare AI Gateway Nothing published about running behind your gateway.
- TrueFoundry AI Gateway Nothing published about running behind your gateway.
- Groq Nothing published about running behind your gateway.
- Together AI Nothing published about running behind your gateway.
Cloud IAM I already run
10 fit · 1 with a caveat · 9 noCan I authenticate with my existing cloud identity instead of another API key?
Fits10
- Braintrust Gateway Reuses IAM roles, workload identity or managed identities.
- Kong AI Gateway Reuses IAM roles, workload identity or managed identities.
- Apache APISIX AI Gateway Reuses IAM roles, workload identity or managed identities.
- Bifrost Reuses IAM roles, workload identity or managed identities.
- LiteLLM Reuses IAM roles, workload identity or managed identities.
- Amazon Bedrock Reuses IAM roles, workload identity or managed identities.
- Azure AI Foundry Reuses IAM roles, workload identity or managed identities.
- Google Vertex AI Reuses IAM roles, workload identity or managed identities.
- TrueFoundry AI Gateway Reuses IAM roles, workload identity or managed identities.
- Fireworks AI Reuses IAM roles, workload identity or managed identities.
With a caveat1
- Vercel AI Gateway Its own OIDC token, not your cloud IAM.
No9
- OpenRouter Static upstream credentials only. Your calls to it still use its own key.
- Requesty No identity integration published.
- Helicone No identity integration published.
- Orq.ai Router Static upstream credentials only. Your calls to it still use its own key.
- Portkey No identity integration published.
- LLM Gateway No identity integration published.
- Cloudflare AI Gateway Static upstream credentials only. Your calls to it still use its own key.
- Groq No identity integration published.
- Together AI No identity integration published.
LangChain or LlamaIndex
14 fit · 5 with a caveat · 1 noI orchestrate in Python. What has documented framework integrations?
Fits14
- OpenRouter LangChain, LlamaIndex
- Requesty LangChain, LlamaIndex
- Braintrust Gateway LangChain, LlamaIndex
- Portkey LangChain, LlamaIndex
- Vercel AI Gateway LangChain, LlamaIndex
- Apache APISIX AI Gateway LangChain, LangGraph, LlamaIndex
- LiteLLM LangChain, LlamaIndex
- Amazon Bedrock LlamaIndex, Strands
- Azure AI Foundry LangChain, LangGraph, LlamaIndex
- Google Vertex AI LangChain, LlamaIndex
- TrueFoundry AI Gateway LangChain, LangGraph, LlamaIndex
- Fireworks AI LangChain, LlamaIndex, CrewAI, PydanticAI, Strands
- Groq LangChain, LlamaIndex
- Together AI LangChain, LlamaIndex
With a caveat5
- Helicone LlamaIndex only.
- Kong AI Gateway LangChain only.
- Orq.ai Router LlamaIndex only.
- Bifrost LangChain only.
- Cloudflare AI Gateway LlamaIndex only.
No1
- LLM Gateway No framework integration documented.
MCP servers to govern
7 fit · 12 with a caveat · 1 noI have MCP servers and need one place to control access to them.
Only an MCP gateway answers this. A hosted MCP server or MCP tool support in a completion API is a different product shape wearing the same acronym.
Fits7
- Requesty Acts as an MCP gateway or registry.
- Kong AI Gateway Acts as an MCP gateway or registry.
- Portkey Acts as an MCP gateway or registry.
- Apache APISIX AI Gateway Acts as an MCP gateway or registry.
- Bifrost Acts as an MCP gateway or registry.
- LiteLLM Acts as an MCP gateway or registry.
- TrueFoundry AI Gateway Acts as an MCP gateway or registry.
With a caveat12
- OpenRouter Hosted MCP server — governs nothing on your side.
- Braintrust Gateway Hosted MCP server — governs nothing on your side.
- Helicone MCP client tooling — governs nothing on your side.
- Orq.ai Router Hosted MCP server — governs nothing on your side.
- Vercel AI Gateway MCP client tooling — governs nothing on your side.
- LLM Gateway Hosted MCP server — governs nothing on your side.
- Amazon Bedrock MCP tools in the API — governs nothing on your side.
- Azure AI Foundry MCP tools in the API — governs nothing on your side.
- Cloudflare AI Gateway MCP tools in the API — governs nothing on your side.
- Google Vertex AI MCP tools in the API — governs nothing on your side.
- Fireworks AI MCP tools in the API — governs nothing on your side.
- Groq MCP tools in the API — governs nothing on your side.
No1
- Together AI Mentioned in docs only
Nothing — plain Node or Python
9 fit · 5 with a caveat · 6 noI want to write code and call something. No console, no cluster.
Fits9
- OpenRouter Change one base URL.
- Requesty Change one base URL.
- Braintrust Gateway Change one base URL.
- Helicone Change one base URL.
- Vercel AI Gateway Install a package.
- LLM Gateway Change one base URL.
- Fireworks AI Install a package.
- Groq Install a package.
- Together AI Install a package.
With a caveat5
- Kong AI Gateway You have to run a process locally before any call works.
- Portkey Install a package, but an upstream provider key is a prerequisite.
- Apache APISIX AI Gateway You have to run a process locally before any call works.
- Bifrost You have to run a process locally before any call works.
- LiteLLM You have to run a process locally before any call works.
No6
- Orq.ai Router A cloud console or resource has to exist before your first call.
- Amazon Bedrock A cloud console or resource has to exist before your first call.
- Azure AI Foundry A cloud console or resource has to exist before your first call.
- Cloudflare AI Gateway A cloud console or resource has to exist before your first call.
- Google Vertex AI A cloud console or resource has to exist before your first call.
- TrueFoundry AI Gateway A cloud console or resource has to exist before your first call.
A note on what “no” means here
Almost every “no” on this page is silence in the vendor’s documentation rather than a refused feature. A product with no published Workers guidance may well work fine from a Worker — it is a fetch call — but nobody has committed to that in writing, so verifying it is your job rather than theirs. That distinction is worth more when you are choosing than a confident tick would be.
Reading it the other way round: each product’s own page carries the same checklist scored for that one product, alongside the evidence links behind every answer. See the full list, or answer a few questions and let the site narrow it for you.