Hacker News new | past | comments | ask | show | jobs | submit login
Launch HN: Helicone.ai (YC W23) – Open-source logging for OpenAI
166 points by justintorre75 on March 23, 2023 | hide | past | favorite | 72 comments
Hi HN - Justin, Scott, and Barak here. We're excited to introduce Helicone (https://www.helicone.ai) an open-source logging solution for OpenAi applications. Helicone's one-line integration logs the prompts, completions, latencies, and costs of your OpenAI requests. It currently works with GPT, and can be integrated with one line of code. There’s a demo at https://www.helicone.ai/video.

Helicone's core technology is a proxy that routes all your OpenAI requests through our edge-deployed Cloudflare Workers. These workers are incredibly reliable and cause no discernible latency impact in production environments. As a proxy, we offer more than just observability: we provide caching and prompt formatting, and we'll soon add user rate limiting and model provider back off to make sure your app is still up when OpenAI is down.

Our web application then provides insights into key metrics, such as which users are disproportionately driving costs and what is the token usage broken down by prompts. You can filter this data based on custom logic and export it to other destinations.

Getting started with Helicone is quick and easy, regardless of the OpenAI SDK you use. Our proxy-based solution does not require a third party package—simply change your request's base URL from https://api.openai.com/v1 to https://oai.hconeai.com/v1. Helicone can be integrated with LangChain, LLama Index, and all other OpenAI native libraries. (https://docs.helicone.ai/quickstart/integrate-in-one-line-of...)

We have exciting new features coming up, one of which is an API to log user feedback. For instance, if you're developing a tool like GitHub Copilot, you can log when a user accepted or rejected a suggestion. Helicone will then aggregate your result quality into metrics and make finetuning suggestions for when you can save costs or improve performance.

Before launching Helicone, we developed several projects with GPT-3, including airapbattle.com, tabletalk.ai, and dreamsubmarine.com. For each project, we used a beta version of Helicone which gave us instant visibility into user engagement and result quality issues. As we talked to more builders and companies, we realized they were spending too much time building in-house solutions like this and that existing analytics products were not tailored to inference endpoints like GPT-3.

Helicone is developed under the Common Clause V1.0 w/ Apache 2.0 license so that you can use Helicone within your own infrastructure. If you do not want to self-host, we provide a hosted solution with 1k requests free per month to try our product. If you exceed that we offer a paid subscription as well, and you can view our pricing at https://www.helicone.ai/pricing.

We're thrilled to introduce Helicone to the HackerNews community and would love to hear your thoughts, ideas, and experiences related to LLM logging and analytics. We're eager to engage in meaningful discussions, so please don't hesitate to share your insights and feedback with us!




I have _specifically_ thought of writing something just like this, so it's awesome to see it!

One thing I would really like to be storing with my requests is the template and parameters that created the concrete prompt. (This gets a little confusing with ChatGPT APIs, since the prompt is a sequence of messages.) Custom Properties allow a little metadata, but not a big blob like a template. I see there's a way to have Helicone do the template substitution, but I don't want that, I have very particular templating desires. But I _do_ want to be able to understand how the prompt was constructed. There's probably some danger on the client side that I would send data that did not inform the prompt construction, and balloon storage or cause privacy issues, so there is some danger to this feature.

Backoffs and other rate limiting sounds great. It would be great to put in a maximum cost for a user or app and then have the proxy block the user once that was reached as a kind of firewall of overuse.

Note your homepage doesn't have a <title> tag.


This is so great! Thanks for the feedback. This is something we have thought a lot about, and really want to make sure we are adding the "correct" features that don't disrupt the user experience.

I kicked off a thread here https://github.com/Helicone/helicone/discussions/164 and would love your input!

P.S. that for the <title> note


I'll note the jump in cost from Free to Starter will keep me from using this for my own projects, 1000 requests just isn't that much, and if I'm doing something more serious then 50,000 requests also doesn't seem like a lot.


stay tuned for updated pricing


Congrats on the launch. Like a lot of the people right now, am doing some side project with OpenAI + LangChain and immediately got some value out of this, specifically:

* When your chains get long/complex enough in LangChain, it's really hard to understand from debug output what's final prompt that actually being sent, how much it costs, or catching runaway agents. This pretty much solves that for me.

As a "prompt developer", one thing that'd be incredibly useful is a way to see/export all of my prompts and responses over time to help me tune them (basically a "Prompt" button in the left nav).

Congrats on the launch. So nice to see a tool in the space that lets you get up and running in 4 minutes.


Congrats the on launch! I noticed you are referring to the project as open source while using the commons clause, which isn't typically considered an open source license, so I raised this via the discussions of the GitHub repo [1].

[1]: https://github.com/Helicone/helicone/discussions/165


Here's another open source alternative: https://github.com/6/openai-caching-proxy-worker


Aaand another even simpler one: https://github.com/egoist/openai-proxy

This one uses Reflare under the hood: https://github.com/xiaoyang-sde/reflare


These are great! we looked into using reflare as well, however reflare does not support streaming at the moment. We are looking to give back to the community and add it for them.

I opened this issue up last week https://github.com/xiaoyang-sde/reflare/issues/443


It appears that the author has responded to the issue, but you haven't replied yet.


Wow, I had no idea GH allowed single character usernames; GL as well https://gitlab.com/0


Y'all committed an env file https://github.com/Helicone/helicone/pull/136/files O___O


Yeah this was added by accident, luckily we have pretty good management of local development and production testing, so these keys were not production keys


lgtm.

A global .gitignore is one of the first things I install on a new machine - and then I never think of it again.

  $ git config --global core.excludesfile ~/.gitignore


Oh wow, I didn't know a global .gitignore was possible.


TIL.


I mean they launched a YC-funded company that will soon be replaced by a "Usage" dashboard and some admin settings over at OpenAI.


You're not wrong, but surely OpenAI can't do everything and maybe they can stay ahead on features long enough to continue to be higher value?


It's a ritual at my prev workplace. They didn't do anything even after i warned about it. Of course they got hacked..


Lol, has their SB creds. Someone could do a bunch of stuff with that


These appear to be local credentials (supabase init, supabase start), but I'll reach out to the founders now to make sure everything is secure on their Production database/APIs. We are a GitHub secret scanning partner [0], so hopefully this was caught early.

---

For any other founders reading this, it's recommended to add a `SECURITY.md` to your repo before doing a ShowHN/LaunchHN. This can be exposed in your `.well-known` folder (eg: https://supabase.com/.well-known/security.txt). This will help with responsible disclosures.

[0] GitHub secret scanning: https://github.blog/changelog/2022-03-28-supabase-is-now-a-g...


> Do not reveal the problem to others until it has been resolved,

sorry, probably shouldn't have pointed that out. noted for future reference.

aside: big fan of Supabase, Paul! it's a pleasure using it!


This is exactly right, thanks a bunch for checking. Also, thanks for the note! We will add a SECURITY.md


The entire company is based around the idea of providing metrics for one closed-source platform's API? Or is the "OpenAI application observability" just one part of what the company does? Otherwise it seems like taking the "all eggs in one basket" to the extreme.


OpenAI is just the start to capture most of the market and iterate on some ideas first :). We are adding multiple providers soon. ~Stay tuned~


What providers apart from OpenAI exist?



"all eggs in one basket" to the extreme.

You couldn't describe better what a startup must do.


To be fair, there are companies based on providing metrics for AWS, e.g. Vantage, so it's not surprising to see VCs making bets in this area


AWS provides services and APIs to support that specific usage. This service is MITM'ing another company's API. I'm pretty sure AWS wouldn't tolerate that sort of thing.


One could think about it as consulting, but productized, on top of a service. Something like AWS is mind numbingly complicated. Amazon has solutions engineers but there's also a big 3rd party market. And some part of that market doesn't need fully bespoke solutions but can be well served by a SaaS that sits on top (of course entailing risk, but it's all risky).

OpenAI isn't there yet, personally I'm bearish on generative AI and openAIs dominance, but they're the major player, and there's a version of the future where they dominate the landscape with a complex enterprise offering like AWS that leaves room for intermediaries. So it's not a bad idea to include bets on people who are building these intermediary products as part of a vc portfolio.


(my understanding might in incorrect but) is it not how cloudflare prevents DDOS attack?


AWS is a whole other beast than the few dozen REST and SSE endpoints at OpenAI.


"Solve one pain point for OpenAI's users, get some nice traction, get acquired by OpenAI in a couple of months" is probably the plan here. Not a bad one to turn a quick profit.


Do you think OpenAI is not logging their users' interactions already?


They certainly aren't exposing those logs back to the customer. There doesn't seem to be a way to see anything other than "total cost per day" in the dashboard. I like that Helicone allows me to see each request and export them all, so that I can use that to build my fine-tuning dataset.


True, but why would they acquire a company for something they already have: the users and the logs. The tech is trivial.


It seems pretty brittle to me also. And their solution is to actually _proxy_ OAI's API wholesale which doesn't seem like a very good idea and might not even be permitted.


I guess they are betting that openAI will be such a big platform that many companies will need this capability. Not a bad idea I guess


It's a big basket.


Does this run afoul of OpenAI's terms of of service in any way? Using a commercial proxy/broker like this to access their API services instead of using directly.


More direct concern: PII leaking from end-users will be a problem if the usage of this service isn't made clear and consented with from each individual. I really think the responsible thing by providers like OP is to remind their customers about this.

If you have end-users in for example EU then integrating a service like Helicone would usually require new informed consent to be legal.


I think you could dispense with this issue by just adding Helicone to your list of subprocessors in your GDPR disclosure.


We see a lot of applications on the market essentially "proxying" traffic to OpenAI, moving to a non-proxy implementation is not too hard if they have any issues with us doing this.


I mean from your users' perspective, not yours. Since it's their OpenAI access that will get revoked for violating the TOS.


The reality is that virtually all of the public AI tech available right now is from OpenAI. Over time as more models are commercialised and generally available, it's probable that Helicone will serve them too.

Congrats Justin and team! Excited for you.


Couldn't agree more. Thanks! :)


Could be cool if you offer quota enforcement. So if something goes rogue on the application side, your layer could not only offer observability but a ceiling to protect against unexpected cost overruns.


Wow so cool! Does this act kind of like a logger and would we be able to have access to the logs later on, or should we bring our own logger as well?

(Also just curious, are you guys just using D1 or KV under the hood?)


Thanks! :D No need to bring your own logger, just change one line of code and then all your logs are in Helicone.

We are using CF Worker cache under the hood https://developers.cloudflare.com/workers/runtime-apis/cache...

We might use D1 for some of our other features like rate limiting.


What is the market for this solution? Whose pain point(s) are you solving? And what stops OpenAI from "Sherlocking"* you, i.e., making whatever you're building a free included feature and extinguishing the market?

*Or, to use a more modern analogy, "Evernoting"


Great question! The short answer is nothing! We just have to out pace them and make sure we develop a better product experience.

Two key advantages that we have over OpenAI. 1. We are open source and are trying to build a community of developers that can out run any single LLM provider. 2. We are completely provider agnostic, which allows us to aggregate and share common features across tools. (One analogy we like to use, is we want to emulate how dbt is database agonistic, and Helicone will grow to be provider agnostic)


Great answer


This is great, however, I am concerned that the various AI 'ecosystem' of all the bolt-on, add-on, plug-ins etc... will be like a billion services all looking for payment - and any complex startup that needs a bunch of these services to build their own service/product/platform, it will be like the current state of streaming services.

So, youll be trying to manage a ton of SLAs, contracts, payment requirements, limits on service access that may be out of your budget to pay for all the various services, API calls, etc.

This is going to be an interesting cluster....

So we need a company thats a single service to access all available AI connects and the multiple billing channels.

However, then you have that as a single POF


So route traffic through you so that you can monetize the insights from the data??


LLM infrastructure is the spot to be right now for startups.


Happy Helicone customer here. It's a dead simple setup. It's great to have the extra charts and logging to debug issues and make sure all is running well.

Congrats to the team!


Thanks! Please reach out to the team on discord if you have any questions


congrats! Helicone provides one of the biggest missing pieces from the the AI tool dev experience today, thanks for building this and sharing it with the rest of us!


Congrats! We've been happy users of Helicone for the past few months--it literally helped us solve a bug with OpenAI's API where we didn't know why requests were failing and we failed to log some of their responses. Helicone helped us debug that it was a token limit issue really quickly, especially since the logging around hasn't been great.

Love how easy it was to integrate too--just one line to swap out the OpenAI API with theirs.


Thanks! Glad Helicone was able to help your workflow.


Perfect! I was just today thinking about how I need to build up my own data set and should be logging all of my transcripts. This is exactly what I wanted.

I want to gather up my chat transcripts, then identify poor experiences in the chat, and then use that to guide fine-tuning. I don't believe that OpenAI actually provides anything to enable this as part of their platform, right?


Awesome.

Is there a consumer version of this?

Like an alternative ChatGPT client or chrome extension that will save my prompts/conversations, tell me which ones I liked more and let me search through them?


From a VC perspective, this sounds like easy money, since it'll probably be acquired by OpenAI or Microsoft if it succeeds just a little.


This is exactly what we needed. We were one day away from an internal build out so the timing couldn't be better.


Hmm, so to integrate I have to basically send my api key to you on every request? Not great


Are you going to be peeking at the data for your own use or is that off limits somehow?


Is data encrypted in such a way so that your company cannot read conversations?


onboarded and started using it in literally 2 minutes. nice


we're glad you were able to get onboarded so quickly! we're always happy to incorporate any feedback you may have on the onboarding experience


This is awesome guys!


Congrats on the launch!




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: