Hacker News new | past | comments | ask | show | jobs | submit login
Launch HN: Release (YC W20) – Orchestrate AI Infrastructure and Applications
73 points by tommy_mcclung 38 days ago | hide | past | favorite | 39 comments
Hello Hacker News! We’re Erik, Tommy, and David, the founders of Release (https://release.ai/). We launched on HN in 2020 (https://news.ycombinator.com/item?id=22486031) after leaving TrueCar, where we managed a 300 person development team. Our original focus was making staging environments easier with ephemeral environments, but along the way AI applications started to emerge as an important and critical component of distributed applications. As we talked to customers using our original product, we realized we had built the underlying platform needed to address the needs of orchestrating AI applications and infrastructure. So here we are and we’re excited to share Release.ai with HN.

Here’s a video showcasing the platform and demonstrating how to easily manage new data and changes using the RAG stack of your choice: https://www.youtube.com/watch?v=-OdWRxMX1iA

If you want to try release.ai out, we’re offering a sandbox account with limited free GPU cycles so you can play around and get a feel for Release.ai: https://release.ai. We suggest playing around with some of the RAG AI templates and adding custom workflows like in the demo video. The sandbox comes with 5 free compute hours on an Amazon g5.2xlarge instance (A10 with 24GB VRAM, 8vCPUs and 32GB). You will also get 16 GB and 4vCPUs for cpu workloads such as web servers. You will be able to run an inference engine plus things like an api server, etc.

After the sandbox expires, you can switch to our free plan, which requires a credit card and associating an AWS/GCP account with Release to manage the compute in your cloud account. The free account provides 100 free managed environment hours a month. If you never go over, you never pay us anything. If you do, our pricing is here: https://release.com/pricing.

For those that like to read more, here’s the deeper background.

It’s clear that open source AI and AI privacy are going to be big. Yes, many developers are going to choose SaaS offerings like OpenAI to build their AI applications, but as open source frameworks and models improve, we’re seeing a shift to open source running on cloud. Security and privacy is a top concern of companies leveraging these SaaS solutions, which forces them to look at running infrastructure themselves. That’s where we hope to come in: we’ve built Release.ai so all your data, models and infrastructure stay in your cloud account and open source frameworks are first class citizens.

Orchestration - Integrating AI applications into a software development workflow and orchestrating their lifecycle is a new and different challenge than traditional web application development. Release also makes it possible to manage and integrate your web and AI apps using a single application and methodology.

To make orchestrating AI applications easier, we built a workflow engine that can create the complex workflows that AI applications require. For example, you can automate the redeployment of an AI inference server easily when underlying data changes using webhooks and our workflow engine.

Cost and expertise - Managing and scaling the hardware required to run AI workloads is hard and can be incredibly expensive. Release.ai lets you manage GPU compute resources across multiple clouds with different instance/node groups for various jobs within a single admin interface. We use K8s under the covers to pull this off. With over 5 years of building and running K8s infrastructure our customers have told us this is how it should be done.

Getting started with AI frameworks is time consuming and requires some pretty in-depth expertise. We built out a library of AI templates (https://docs.release.com/release.ai/release.ai-templates) using our Application Template format (which is kind of a super docker-compose: https://docs.release.com/reference-documentation/application...) for common open source frameworks to make it easy to get started developing AI applications. Setting up and getting these frameworks running is a hassle, so we made it one click to launch and deploy.

We currently have over 20 templates including temples for RAG applications, fine tuning and useful tools like Juypter notebooks, Promptfoo, etc. We worked closely with Docker and Nvidia to support their frameworks: GenAI and Nvidia NEMO/Nims. We plan to launch community templates soon after launch. If you have suggestions for more templates we should support, please let us know in the comments.

We’re thrilled to share Release.ai with you and would love to get your feedback. We hope you’ll try it out, and please let us know what you think!




I've noticed that while a bunch of developers have played with LLM's for toy projects, few seem to have any actual experience taking it to prod in front of real users. I’ve personally had to do so for a few startups, and it's like trying to nail Jell-O to a tree. Every random thing you change, from prompts to models, yields massively different/unpredictable results.

I think because of this, a bunch of companies/tools have tried to hop in this space and promised the world, but often times people are best served by just hitting OpenAI/GPT directly, and jiggling the results until they get what they want. If you're not comfortable doing that, there are even companies that do that for you, so you can just focus on the prompt itself.

So that being said, help me understand why I should be adding this whole system/process to my workflow, versus just hitting OpenAI/Anthropic/Google directly?


You're right - hitting OpenAI/Anthropic/Google directly is often the quickest way to get started, and for many simple applications, it might be all you need. However, Release.ai addresses the needs of companies that require more control, customization, and scalability in their AI systems.

Release.ai isn't about replacing the big players but about giving you options. It's for when you need more than a generic API call but don't want to build an entire ML infrastructure from scratch. You can build exactly what you need without getting a Ph.D. in machine learning or becoming a DevOps expert.


define more control, define more scalability and also define more customization


I'm in the process of rolling out an LLM to a user facing feature and it's difficult. The scaling is not obvious, and the quality fluctuates even with Llama-3.1 (8B) when compared to GPT-4o. We're probably going with 4o since the JSON return works much more reliable, it follows instructions for text generation more directly, etc.


This looks awesome, getting started with AI development is daunting and I really like how this focuses on integrating with a bunch of open source frameworks and then deploying them into your own cloud (I always prefer to run the infrastructure, it feels weird to rely on something that's a complete black box).

The sandbox environment with free GPU hours is a cool way to try things out without a big commitment too. It's nice seeing a product that genuinely seems to address the practical challenges of AI deployment. Looking forward to seeing how the platform develops!


Thanks! Hopefully, the sandbox allows people to try out some things and see how Release works. Release become most powerful when you are deploying into your own infra and mixing and matching your web apps and AI apps.


Super interesting you guys have been working on this since 2020 if I'm reading the post title correctly? Would love to know the iterations you've gone through.


It would be hard to count. Basically every customer we had on Release 1.0 was an iteration towards this. If you go to https://release.com it's there and we have many happy customers using it. There are definitely challenges with that business, however, that made us look for easier ways to get people using the platform we had built. People looking to do ephemeral environments have a lift to move their environment definitions into Release and it's a work effort that needs prioritization. Release.ai is built on the same platform and because AI frameworks are more turn key than bespoke software stacks companies have been building on, we believe Release.ai will be easier to adopt. In the long run AI applications and traditional web applications are going to merge and we think we're the platform that will do that in the long run. Long story short, hundreds of iterations.


This is very cool! I love seeing tooling targeting inference. I feel like stable diffusion and LLAMA have to be the primary use cases for these types of services. DALL-E is super lacking and GPT does actually start to get pretty expensive once you are using it in production.


This looks cool, but I'm a little confused about the pricing model. It sounds like I'm paying you for every hour my jobs are running on my own infrastructure if I'm reading it right. That seems like a really odd way to price things if true.


It's not a unique model - that's what e.g. Databricks does too. You pay for the resources, which means you can deploy them with whatever security model you require (good for enterprise, with all the firewalls and secrets and security reviews your heart desires), but you also pay for the management layer that Databricks offers. In Databricks' case you're basically doubling the infrastructure cost per hour.

Data and compute stay in your own tenant.

Edit: confirmed - look at that enterprise tier. If you want SSO & RBAC you click that button and pay $5k/month minimum. Definitely an enterprise play and the pricing model and approach to security will make sense to those customers.


Fair enough. I'm just not familiar with that pricing model, but that somewhat makes sense.


Yeah, I hear you and the confusion. We decided on this as a management fee that only gets charged when you’re using the environment. It was the best tradeoff we could come up with but it definitely isn’t perfect. Any ideas on how we can make this clearer or better?


How does this compare to managed offerings like Google Gemini and AWS Bedrock? Thanks in advance and congratulations on the new product!!


Release.ai gives you complete control over your infrastructure and allows you to create the workflows that are best for your particular problem. You can go as deep as you would like using the open source AI tools available to you without restrictions.

Release.ai is much cheaper than those options even with our pricing on top of your cloud costs.

We are a single pane of glass no matter where your k8s clusters are running. You don't need to learn bedrock, sagemaker, gemini, etc in order to use gpus in different clouds.

Release also makes it very easy to create complete applications with web services and ai services together. No other platform allows you to do both easily.

We are a complete platform where you have total control over the software you select and how it needs to be integrated in your development processes.

Thanks for the kind words!


How do you hook up data to an environment? And what data sources do you support (Snowflake/etc?)


We have a way to connect data to an environment through workspaces. The docs are here: https://docs.release.com/reference-documentation/application...

Right now we support s3 and pulling in full github repos. Snowflake is on our roadmap and has been requested by some of our customers already.


This is cool. I'd like to give it a try. Press a button, and get GPU access to build apps on.


Thank you! Let us know how you get on with it!


We might not have made it clear in the post how to signup for the sandbox. Just head to http://release.ai and click on "Start Free Trial".


Are there 2 levels of free usage, where the first sandbox doesn't require Google Cloud or AWS accounts, but after the trial, they are required?


Why do you have such generic name? It will make searching so much harder.


It's funny you ask this because this question has come up a bunch of times. It just maps to our mission and we decided we really wanted to be true to it. Our mission is to help people/companies release their ideas to the world quickly. So it just fit and we decided it was worth the tradeoff for it to be a little harder to search for. In the end if we're successful it will be because we built something people want (i.e. apple.com) vs the name we chose.


Seems pretty vague. Something to do with half self-hosting open source LLMs with a proprietary docker-like template thing.

Am I on the right track?


Not sure I would describe it that way exactly, :). The templates are to get your started. Release itself takes docker compose, docker files, helm, terraform, etc and gives you a way to organize them and a workflow engine to deploy, update and remove every part of the environment. All of the functionality works with the AI templates.

The templates are a good place to start and with our concept of workspaces, you an easily attach data sources and do development with common ai tools and frameworks.

A lot of our customers and others we have talked to don't even know where to start with AI, including RAG. Our templates, infra management and workflow engine make it easy to spin up multiple environments, with varying configs for testing all the way to production.

While a lot of companies are getting into the RAG framework or Fine Tuning framework business, by releasing open source project, it is still very difficult to tie it all together.

Both nvidia and Docker have lamented about this issue to us and this is our attempt to make some of these frameworks easier to use.


Like Together.ai but a bit late to the market I suppose.


Pretty insane that these folks are YC 2020 and just now pivoted again. Shows how hard company building is.


It's not a "pivot" in the traditional sense. But yeah, building companies is probably one of the hardest things one can do. We still run https://release.com and https://release.ai is built on the same platform.


Congrats on launching. Interesting pivot.


Thank you! We believe our standard product and Release.ai work really well together!


Great system, no complaints!!


I'm pretty much exactly your target market. I run a kubernetes, docker, poetry devops hell at an ml startup and was curious how your product helped. You got about 2 minutes of my time. I scanned your website, I have no idea what you do or whether you fix my problem.

Not trying to be negative, but I think there may be a 30 second pitch to be made to people like me that isn't made on your site.


Thanks for the feedback and I agree we need to do a better job messaging the value on the site. How would this have landed for you?

We are an AI orchestration platform that makes deploying open source models and frameworks using Kubernetes and docker simple. We manage the GPU and k8s resources, provide templates to common open source frameworks and an orchestration engine for your AI workflows.


I think at the end of the day, I have a bunch of config files and commands. Showing a quick translation between the spaghetti and your solution would make the most sense. The less language I have to parse the better.

i.e. you have a set of kubernetes CRDs probably managed by a gitops solution -> we do this for you, you have a bunch of different pyproject.toml files that need to be orchestrated into a docker container -> this is what that looks like in release, tbh our terraform isn't a huge bottleneck because so much just ends up going in the kube crds.

maybe have two pitches: 1 for startups getting started and 1 for startups who already have a lot of this figured out (probably your bigger target audience if you have a strong pitch)


Awesome feedback, thank you!!


congrats!!


Thanks!


What has been the feedback from early users regarding the ease of transitioning from your original product focus to the current AI orchestration platform?


They didn't have to transition. https://release.com is still alive and customers are happily using it.




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

Search: