Prompt capture/logging

[livid escarp]

Hey guys, we run a BYOK service with OpenAI and I'd love to expand to more models using OpenRouter. I'm curious whether end users will have any way to intercept and capture our prompts if they use their own OpenRouter accounts..?

Thanks!

[livid escarp]

Any thoughts on this guys?

[calm shard]

I think there is a prompt/completion logging option in the settings? try it yourself and see if you can read the prompts

[livid escarp]

I played with that but didn’t see a way to access the logs myself, maybe I missed something..?

Its also a bit unclear to me what that setting does, ie is it for the-end user or for openrouter/providers? It seems to be a privacy setting, so I assumed the latter…

It would still be great to get confirmation from the team about whether end-user can access a log of prompts..?

[lone dust]

Hey there, are you calling the API from a server backend or from a frontend?

We're not exposing prompt or completion to users, unless your prompt lives on the front end

But otherwise your end user shouldn't be able to see it

Are you using OAuth2?

[digital flume]

The logging option in your preferences only control if you allow OpenRouter to classify your prompts for the model tags.

It is impossible to retrieve a prompt again once it is sent as a reply.

[livid escarp]

It will always be backend

[livid escarp]

Okay good to know that

[livid escarp]

Okay so sounds like there is no way for users to log/capture api calls..?

Do you allow (or plan to allow) end-users to add/register their own models? Ie that would give someone the ability to capture prompts…

[digital flume]

You can of course build you own middleware and send the prompt through there.

[livid escarp]

What do you mean by this?

[digital flume]

Think of a proxy. Basically the same but in small and only for you app what OpenRouter does.

[livid escarp]

We’re a bring-your-own-key service and want to stay like that, so we need the ability for users to connect their own api keys.

We’re OpenAI only for the moment and want to expand to others but rather than building against multiple apis found ourselves here 🙂

[digital flume]

Do you use OpenAI threads to store an retrieve user prompts? Otherwise I don't know how that should work on their side?

[livid escarp]

Not sure I understand the relevance here sorry 🙂

We use the simple openai chat endpoint and our users cant see our prompts - we’d like to use OpenRouter but can only do so if our prompts won’t leak…

[digital flume]

I thought the goal was to retrieve past prompts on purpose?

[livid escarp]

No the opposite, we want ti prevent our prompts leaking

[digital flume]

As I said, it is impossible to retrieve a prompt a second time, even with logging turned on.

[livid escarp]

Sure I got that part 🙂

The second question was whether OpenRouter allows people to register/use their own models, exposing prompt capture that way.?

[digital flume]

No user models. OpenRouter routes to popular openly accessible commercial LLM endpoints only.

[calm shard]

are the prompts deleted after that? I was under the impression that it's more long-term storage and retention as well
I'm much less hesitant to help OpenRouter in this way so I'm wondering if that's the case

[lone dust]

by "that," you mean after we're done with the classification?

if logging is allowed, the prompts/completions are anonymized before storage, and we throw it all in a bucket atm

[calm shard]

yep! okay so my initial impression was correct I think

about there being long-term retention (albeit anonymized)