Hi there HN, want to share a project in the works for five years. After working in Microsoft Research and then RStudio, I left to start Hal9 to build a WebML AI platform. Over five years I've learned a lot and pivoted many times, but we kept building the core infrastructure that runs docker containers. This infra became Cloudish and we just launched it.
Cloud providers today are designed to have a human in many steps, starting with sign-up and even talking to support or sales people to increase your quotas. In addition, most send you an invoice at the end of the month for whatever you consume. This design causes friction at best and a $10K surprise bill at most.
Cloudish starts with the agent, not the human. Agents running in Claude, ChatGPT, Grok, Cursor, etc. can simply mint their own key by calling an API and deploy a docker container. Cloudish provides the build system, multiple instance types, logs for the build, an image registry, runtime logs, volume persistence and basic OIDC authentication. We achieve this by making the https://cloudish.ai/skill.md file the main interface to agents with an optional website https://www.cloudish.ai/ for humans to claim their projects, manage them and buy credits. Currently we top up with Stripe but have 402 payments in the not-so-distant roadmap.
Our main differentiator is ease of use for agents. We want to enable agents to deploy fully autonomously, so we created an open source benchmark to measure if agents can deploy to cloud providers without user intervention (https://github.com/cloudishai/ai-deploy-bench) and we will keep measuring progress towards this over time.
We've been dogfooding this with various projects that run in Cloudish, here are some of our favorite ones:
- htttps://www.cloudish.ai: The website itself runs on cloudish.ai
- htttps://demo.cloudish.app: This one shows how you can build and deploy a simplified Lovable clone that uses a docker sandbox to run Claude Code and then deploy the websites back to Cloudish
- htttps://awesomedataviz.com: A list of data viz projects
- htttps://puzzle.cloudish.app: Uses computer vision to solve puzzles
- htttps://whatislocalhost.com: Explains what is localhost
- htttps://lightpaths.cloudish.app: An R ray tracing demo created with rayrender
- htttps://neurosciencebook.com: Under development, chapter 1 written
javierluraschi•4m ago
Cloud providers today are designed to have a human in many steps, starting with sign-up and even talking to support or sales people to increase your quotas. In addition, most send you an invoice at the end of the month for whatever you consume. This design causes friction at best and a $10K surprise bill at most.
Cloudish starts with the agent, not the human. Agents running in Claude, ChatGPT, Grok, Cursor, etc. can simply mint their own key by calling an API and deploy a docker container. Cloudish provides the build system, multiple instance types, logs for the build, an image registry, runtime logs, volume persistence and basic OIDC authentication. We achieve this by making the https://cloudish.ai/skill.md file the main interface to agents with an optional website https://www.cloudish.ai/ for humans to claim their projects, manage them and buy credits. Currently we top up with Stripe but have 402 payments in the not-so-distant roadmap.
Our main differentiator is ease of use for agents. We want to enable agents to deploy fully autonomously, so we created an open source benchmark to measure if agents can deploy to cloud providers without user intervention (https://github.com/cloudishai/ai-deploy-bench) and we will keep measuring progress towards this over time.
We've been dogfooding this with various projects that run in Cloudish, here are some of our favorite ones:
- htttps://www.cloudish.ai: The website itself runs on cloudish.ai - htttps://demo.cloudish.app: This one shows how you can build and deploy a simplified Lovable clone that uses a docker sandbox to run Claude Code and then deploy the websites back to Cloudish - htttps://awesomedataviz.com: A list of data viz projects - htttps://puzzle.cloudish.app: Uses computer vision to solve puzzles - htttps://whatislocalhost.com: Explains what is localhost - htttps://lightpaths.cloudish.app: An R ray tracing demo created with rayrender - htttps://neurosciencebook.com: Under development, chapter 1 written