Does AI Gateway return any kind of info when using the Universal Endpoint to know which model was su

OOblomki Does AI Gateway return any kind of info when using the Universal Endpoint to kno...

K

Kathy•6/23/25, 10:30 PM

https://developers.cloudflare.com/ai-gateway/configuration/fallbacks/#response-headercf-aig-step

Cloudflare Docs

Fallbacks

Specify model or provider fallbacks with your Universal endpoint to handle request failures and ensure reliability.

D

dave•6/24/25, 4:35 PM

Anthropic Claude Sonnet 4 and Opus 4 pricing are still very wrong, they aren't taking into account cache related tokens at all.

T

Tait•6/24/25, 8:55 PM

I was trying to https://github.com/googleapis/js-genai with AI Gateway. Is that supported? My inital take after fighting with this for a while is that I should use the deprecated google library https://github.com/google-gemini/deprecated-generative-ai-js

Am I completely wrong? I hope so. I'm trying to use gemini to process short videos and I'd love to track my codst.

GitHub

GitHub - googleapis/js-genai: TypeScript/JavaScript SDK for Gemini ...

TypeScript/JavaScript SDK for Gemini and Vertex AI. - googleapis/js-genai

GitHub

GitHub - google-gemini/deprecated-generative-ai-js: This SDK is now...

This SDK is now deprecated, use the new unified Google GenAI SDK. - google-gemini/deprecated-generative-ai-js

P

piffie•6/25/25, 7:42 AM

Is there a better Frontend for the AI-gateway? where i can see maybe prompt versions or similar?

TTait I was trying to https://github.com/googleapis/js-genai with AI Gateway. Is that...

A

AngusMa•6/29/25, 3:22 AM

I think AI Gateway is based on REST API?

S

Shados•6/29/25, 5:17 AM

I'm using a gateway with a Worker AI binding (the "basic usage" in section 2 on https://developers.cloudflare.com/ai-gateway/integrations/worker-binding-methods/)

I don't get any error and my app works, but I don't see anything in the gateway's logs.

If I use the REST endpoint to my gateway to make the same request to the same model, then I see log entries. Is that a bug or am I doing something wrong?

If I use this.env.AI.gateway then gateway.run, then it works as I expected, but since the doc and the typescript type explicitely state the gateway feature should be supported directly, I found the behavior surprising.

S

Shados•6/29/25, 5:50 AM

Even then something seems off. Some features like input/output tokens and cost don't register when I go through the

gateway.run

gateway.run

gateway.run

gateway.run function inside workers.

So so far:

using a regular Worker AI binding with the gateway property: doesn't show up in logs at all
Getting the gateway by id from the binding and calling gateway.run: shows up in logs, no token/cost show up
Using REST api via curl: everything works as expected.

(it case it matters, im using the Worker AI models in this case)

Edit: Ok token/cost is not REST api vs gateway binding, its Streaming vs Non-Streaming.

So that only leaves the issue of: Why do I not get logs if I'm using the bindings without gateway.run? Is it because it only logs queries that go through the API and direct bindings bypass that? In which case what features do I actually get, since logs are kind of the big one?

A

ac•6/29/25, 7:12 PM

We are having issues where OpenAI requests are now getting blocked because they are originating from unsupported countries (via AI Gateway). This didn't start happening until this week for us, but it seems to be a new error as of the past few weeks (see https://community.cloudflare.com/t/cloudflare-worker-blocked-by-openai/804750

https://community.cloudflare.com/t/ai-gateway-routes-to-restricted-regions-causing-api-blocks/808571/6).

an anyone at Cloudflare comment on this?

Aac We are having issues where OpenAI requests are now getting blocked because they ...

H

harshil1712•6/30/25, 10:33 AM

forwarding this to the team.

Hharshil1712 forwarding this to the team.

A

ac•6/30/25, 10:37 AM

Thank you! This is a big issue for us.

J

jedi•6/30/25, 12:16 PM

hi guys

J

jedi•6/30/25, 12:16 PM

how can i apply temperature?

J

jedi•6/30/25, 12:16 PM

in the request via api?

V

Vk•7/1/25, 7:40 PM

Do you support custom domain for AI Gateway? Can we use AI gateway to invoke all the worker AI inference points? (Or are there exceptions)?

N

neil•7/3/25, 2:04 AM

hey! currently seeing issues (not sdk - appears to be cf's backend issue) with Universal Endpoint's Authorization header forwarding FOR ALL PROVIDERS

only for Authenticated Gateways. disabling Authenticated Gateway fixes the issue.

_Example__
Request:

curl https://gateway.ai.cloudflare.com/v1/[account_id]/[gateway-id] \
  --header 'Content-Type: application/json' \
  --header 'cf-aig-authorization: Bearer [cf-aig-token]' \
  --data '[
    {
      "provider": "openai",
      "endpoint": "chat/completions",
      "headers": {
        "Authorization": "Bearer [openai_api_key]",
      },
      "query": {
        "model": "gpt-4o-mini",
        "messages": [
          {
            "role": "user",
            "content": "What is Cloudflare?"
          }
        ]
      }
    }
  ]'

curl https://gateway.ai.cloudflare.com/v1/[account_id]/[gateway-id] \
  --header 'Content-Type: application/json' \
  --header 'cf-aig-authorization: Bearer [cf-aig-token]' \
  --data '[
    {
      "provider": "openai",
      "endpoint": "chat/completions",
      "headers": {
        "Authorization": "Bearer [openai_api_key]",
      },
      "query": {
        "model": "gpt-4o-mini",
        "messages": [
          {
            "role": "user",
            "content": "What is Cloudflare?"
          }
        ]
      }
    }
  ]'

curl https://gateway.ai.cloudflare.com/v1/[account_id]/[gateway-id] \
  --header 'Content-Type: application/json' \
  --header 'cf-aig-authorization: Bearer [cf-aig-token]' \
  --data '[
    {
      "provider": "openai",
      "endpoint": "chat/completions",
      "headers": {
        "Authorization": "Bearer [openai_api_key]",
      },
      "query": {
        "model": "gpt-4o-mini",
        "messages": [
          {
            "role": "user",
            "content": "What is Cloudflare?"
          }
        ]
      }
    }
  ]'

curl https://gateway.ai.cloudflare.com/v1/[account_id]/[gateway-id] \
  --header 'Content-Type: application/json' \
  --header 'cf-aig-authorization: Bearer [cf-aig-token]' \
  --data '[
    {
      "provider": "openai",
      "endpoint": "chat/completions",
      "headers": {
        "Authorization": "Bearer [openai_api_key]",
      },
      "query": {
        "model": "gpt-4o-mini",
        "messages": [
          {
            "role": "user",
            "content": "What is Cloudflare?"
          }
        ]
      }
    }
  ]'

Response:

{
    "error": {
        "message": "You didn't provide an API key. You need to provide your API key in an Authorization header using Bearer auth (i.e. Authorization: Bearer YOUR_KEY), or as the password field (with blank username) if you're accessing the API from your browser and are prompted for a username and password. You can obtain an API key from https://platform.openai.com/account/api-keys.",
        "type": "invalid_request_error",
        "param": null,
        "code": null
    }
}

{
    "error": {
        "message": "You didn't provide an API key. You need to provide your API key in an Authorization header using Bearer auth (i.e. Authorization: Bearer YOUR_KEY), or as the password field (with blank username) if you're accessing the API from your browser and are prompted for a username and password. You can obtain an API key from https://platform.openai.com/account/api-keys.",
        "type": "invalid_request_error",
        "param": null,
        "code": null
    }
}

{
    "error": {
        "message": "You didn't provide an API key. You need to provide your API key in an Authorization header using Bearer auth (i.e. Authorization: Bearer YOUR_KEY), or as the password field (with blank username) if you're accessing the API from your browser and are prompted for a username and password. You can obtain an API key from https://platform.openai.com/account/api-keys.",
        "type": "invalid_request_error",
        "param": null,
        "code": null
    }
}

{
    "error": {
        "message": "You didn't provide an API key. You need to provide your API key in an Authorization header using Bearer auth (i.e. Authorization: Bearer YOUR_KEY), or as the password field (with blank username) if you're accessing the API from your browser and are prompted for a username and password. You can obtain an API key from https://platform.openai.com/account/api-keys.",
        "type": "invalid_request_error",
        "param": null,
        "code": null
    }
}

Have tried sending api key with initial request headers, in the body like above, and both. Issue is not just scoped to OpenAI's api - have also tested Vertex AI and issue persists. Anthropic does not have this issue since their api key is supplied in the

x-api-key

x-api-key

x-api-key

x-api-key header.

A

ari•7/3/25, 2:14 AM

We've been seeing something similar. Universal endpoint + Authenticated Gateway. We have OpenAI as primary and Anthropic as secondary, and OpenAI consistently fails with this same error message and then Anthropic succeeds. On one hand, it's a win for the cross-vendor fallback model.

A

ari•7/3/25, 4:08 AM

In the process of trying to debug this, I came across conflicting documentation on where the

authorization

authorization

authorization

authorization should be stored. Is this out-of-date? https://developers.cloudflare.com/ai-gateway/universal/#:~:text=The%20payload%20is%20expecting%20an%20array%20of%20message%2C%20and%20each%20message%20is%20an%20object%20with%20the%20following%20parameters

N

neil•7/3/25, 5:11 AM

seems like it - the worker gateway binding also doesn't have types for this prop

N

neil•7/3/25, 5:12 AM

Screenshot_2025-07-02_at_10.12.23_PM.png

A

ari•7/3/25, 6:47 AM

I did find one call that worked that I wasn't expecting: moving the OpenAI-specific authorization up to the top level, next to the

cf-aig-authorization

cf-aig-authorization

cf-aig-authorization

cf-aig-authorization gateway-level setting. Perhaps this is header configuration hierarchy at work? https://developers.cloudflare.com/ai-gateway/universal/#header-configuration-hierarchy

Screenshot_2025-07-02_at_11.41.20_PM.png

N

neil•7/3/25, 9:05 AM

hm i just tried putting it in the top level header - both with and without it being in the request body

N

neil•7/3/25, 9:05 AM

didn't work for curl at least

N

neil•7/3/25, 9:06 AM

either way, authorization should def be a request level header for universal endpoint

V

Vk•7/3/25, 1:51 PM

Hello folks: Trying to understand AI Gateway better. With the openAI compatible endpoint, does it support all the models and integrations using compatible endpoint or are there specific ones"

N

neil•7/3/25, 8:21 PM

it’s basically a reverse proxy for rest endpoints (primarily nice for centralized logging). you can specifically request a url path with the endpoint body parameter for universal endpoint or just concat to the url path for provider specific ai gateway

N

neil•7/3/25, 8:22 PM

so for example you can request openai’s /chat/completions as well as /responses with { model: “gpt-4.1” | “gpt-4o” } in the request body

C

Cappy Ishihara•7/4/25, 6:32 PM

currently trying to use the CF AI gateway with Zed to run tool calls, seems like it's failing for some reason to call any tools

Ddave Anthropic Claude Sonnet 4 and Opus 4 pricing are still very wrong, they aren't t...

C

Carlos Alberto•7/11/25, 11:12 AM

If you combine deepseek with just a few MCPs you will get a much more powerful model than Anthropic and at a fraction of the cost

CCarlos Alberto If you combine deepseek with just a few MCPs you will get a much more powerful m...

D

dave•7/11/25, 12:17 PM

I’m using Claude Max 20x so my “real” cost for tokens is fixed at $200/m, regardless of how many I use.

Ddave I’m using Claude Max 20x so my “real” cost for tokens is fixed at $200/m, regard...

C

Carlos Alberto•7/11/25, 2:28 PM

Ur spending $2400 a year to use AI ?

CCarlos Alberto Ur spending $2400 a year to use AI ?

D

dave•7/11/25, 2:29 PM

Far more than that.

Ddave Far more than that.

C

Carlos Alberto•7/11/25, 2:31 PM

What is ur use case?

CCarlos Alberto What is ur use case?

D

dave•7/11/25, 2:35 PM

Work mostly.

Ddave Work mostly.

C

Carlos Alberto•7/11/25, 2:41 PM

If ur spending over 3k a year on AI you should probably consider setting up your own local one that is private. I use AI all day and night and haven't spent more than $30 in last 2 years

CCarlos Alberto If ur spending over 3k a year on AI you should probably consider setting up your...

D

dave•7/11/25, 2:55 PM

Getting a local instance of Claude Sonnet 4 would be several million.

C

Carlos Alberto•7/11/25, 6:06 PM

Everything is a wrapper bro. All you need is a unified multimodal api, secure AI gateway, a few MCPs, some rules, an OS ui and you're good to go. Total bill shouldn't cost more than a few hundred bucks. You could deploy at zero cost with a little engineering.

D

dan—1106•7/11/25, 6:39 PM

Hello, I'm looking to attach pdf/xlsx/csv files to a request I'm sending to Google AI Studio via AI Gateway. Is that possible? If it is, would someone mind sharing a basic example? Thanks in advance for your help

Ddan—1106 Hello, I'm looking to attach pdf/xlsx/csv files to a request I'm sending to Goog...

D

dan—1106•7/11/25, 7:05 PM

for folks who come across this later, I attached files to my messages like this:

  const url = await gateway.getUrl();
  const client = new OpenAI({
    apiKey: apiKey, 
    baseURL: `${url}compat`,
  });
  for (const file of files) {
    const base64Content = Buffer.from(file.content).toString("base64");
  const messageContent: Array<any> = [
    {
      type: "text",
      text: fullPrompt,
    },
  ];
    messageContent.push({
      type: "image_url", // Use "image_url" for all files.
      image_url: {
        url: `data:${file.contentType};base64,${base64Content}`,
      },
    });
  }
  const completionPromise = client.chat.completions.create({
    model: model,
    messages: [
      {
        role: "user",
        content: messageContent,
      },
    ],
    response_format: zodResponseFormat(responseSchema, "query_response"),
  });

  const url = await gateway.getUrl();
  const client = new OpenAI({
    apiKey: apiKey, 
    baseURL: `${url}compat`,
  });
  for (const file of files) {
    const base64Content = Buffer.from(file.content).toString("base64");
  const messageContent: Array<any> = [
    {
      type: "text",
      text: fullPrompt,
    },
  ];
    messageContent.push({
      type: "image_url", // Use "image_url" for all files.
      image_url: {
        url: `data:${file.contentType};base64,${base64Content}`,
      },
    });
  }
  const completionPromise = client.chat.completions.create({
    model: model,
    messages: [
      {
        role: "user",
        content: messageContent,
      },
    ],
    response_format: zodResponseFormat(responseSchema, "query_response"),
  });

  const url = await gateway.getUrl();
  const client = new OpenAI({
    apiKey: apiKey, 
    baseURL: `${url}compat`,
  });
  for (const file of files) {
    const base64Content = Buffer.from(file.content).toString("base64");
  const messageContent: Array<any> = [
    {
      type: "text",
      text: fullPrompt,
    },
  ];
    messageContent.push({
      type: "image_url", // Use "image_url" for all files.
      image_url: {
        url: `data:${file.contentType};base64,${base64Content}`,
      },
    });
  }
  const completionPromise = client.chat.completions.create({
    model: model,
    messages: [
      {
        role: "user",
        content: messageContent,
      },
    ],
    response_format: zodResponseFormat(responseSchema, "query_response"),
  });

  const url = await gateway.getUrl();
  const client = new OpenAI({
    apiKey: apiKey, 
    baseURL: `${url}compat`,
  });
  for (const file of files) {
    const base64Content = Buffer.from(file.content).toString("base64");
  const messageContent: Array<any> = [
    {
      type: "text",
      text: fullPrompt,
    },
  ];
    messageContent.push({
      type: "image_url", // Use "image_url" for all files.
      image_url: {
        url: `data:${file.contentType};base64,${base64Content}`,
      },
    });
  }
  const completionPromise = client.chat.completions.create({
    model: model,
    messages: [
      {
        role: "user",
        content: messageContent,
      },
    ],
    response_format: zodResponseFormat(responseSchema, "query_response"),
  });

Ddan—1106 Hello, I'm looking to attach pdf/xlsx/csv files to a request I'm sending to Goog...

C

Carlos Alberto•7/11/25, 7:38 PM

Have you tried Google Sheets Appscript? It's free, serverless, secure and has seamless integration with Google's stack. Cloudflare and Google Chrome both run on V8 edge native. So you're good to go.

CCarlos Alberto Have you tried Google Sheets Appscript? It's free, serverless, secure and has se...

I

Isaac McFadyen•7/11/25, 7:47 PM

Not sure how that's related? They were asking specifically about using AI Gateway + Google AI Studio, not Sheets, and even if they used Appscript they would still need to know how to attach files to requests going through AI Gateway.

IIsaac McFadyen Not sure how that's related? They were asking specifically about using AI Gatewa...

C

Carlos Alberto•7/11/25, 7:56 PM

Appscript solves - cold starts, serverless, costs, login, cybersecurity, storage, speed and tech chaining. All you need is a script. Vanilla javascript can automate anything. Cloudflare worker is an excellent proxy. Gemini 2.5 flash can make an optimized script for you within 3-4 iterations if you give clear instructions.

CCarlos Alberto Appscript solves - cold starts, serverless, costs, login, cybersecurity, storage...

I

Isaac McFadyen•7/11/25, 8:12 PM

They didn't ask about Appscript, cold starts, costs, logins, etc. They asked a specific question about a specific task with a Cloudflare product (AI Gateway), so responding with unrelated commentary is unproductive.

IIsaac McFadyen They didn't ask about Appscript, cold starts, costs, logins, etc. They asked a s...

C

Carlos Alberto•7/11/25, 8:13 PM

Then what's your solution Einstein?

CCarlos Alberto Then what's your solution Einstein?

I

Isaac McFadyen•7/11/25, 8:13 PM

I don't have one because I don't know the answer to their question, but telling them to use Appscript is not the answer.

IIsaac McFadyen I don't have one because I don't know the answer to their question, but telling ...

C

Carlos Alberto•7/11/25, 8:15 PM

Why not?

CCarlos Alberto Why not?

I

Isaac McFadyen•7/11/25, 8:17 PM

Because... it's completely unrelated? It's not the Cloudflare product they asked about and it's not the answer to their (very specific) question.

IIsaac McFadyen Because... it's completely unrelated? It's not the Cloudflare product they asked...

C

Carlos Alberto•7/11/25, 8:18 PM

Is that a fact?

Ddan—1106 for folks who come across this later, I attached files to my messages like this:...

V

Victor•7/11/25, 8:19 PM

Just FYI, there's an attachment limit of 20MB with this format as documented in the Files API. Beyond that (or as an alternative to reuse the same upload for multiple prompts), they recommend using their files API and using the reference (from the returned metadata) to it as the new

image_url

image_url

image_url

image_url in the prompt. Full File API Endpoint documentation

Does AI Gateway return any kind of info when using the Universal Endpoint to know which model was su

Similar Threads

Similar Threads

Similar Threads