Hi HN. The team behind Shuttle have been working on Neptune, an open source AI Platform Engineer that reads your code and automatically wires up the cloud infrastructure.
Neptune is released under the Apache License 2.0.
We built this after seeing more teams rely on AI to write code, yet still needing to manually create and maintain cloud config. The AI could write a service, but it could not understand the system around it.
Neptune tries to solve that, it sits in your IDE (via an MCP server). It analyzes your code's intent (using AST and inference) and automatically updates a deterministic infrastructure specIt analyzes the structure of your project, detects the services and dependencies, and generates the infra configuration needed to run it on the cloud. The goal is not to hide everything behind magic. We do this by generating a transparent `neptune.json` spec. You see exactly what is being built. It's a 'Glass Box,' not a Black Box. Our goal is to make infra work understandable, predictable, and automatable.
Right now it can infer things like databases, queues, API services, and background workers. It proposes a change to your schema. You approve it. Then it applies it. The AI doesn't have root access to your cloud; it has access to the spec. It is early and still rough. There are missing pieces and edge cases. We would love to hear where it breaks for you.
We would love any feedback, issues, ideas, and failures. You can test the beta (5mins) here: https://www.neptune.dev/try-it-now
Neptune is released under the Apache License 2.0.
We built this after seeing more teams rely on AI to write code, yet still needing to manually create and maintain cloud config. The AI could write a service, but it could not understand the system around it.
Neptune tries to solve that, it sits in your IDE (via an MCP server). It analyzes your code's intent (using AST and inference) and automatically updates a deterministic infrastructure specIt analyzes the structure of your project, detects the services and dependencies, and generates the infra configuration needed to run it on the cloud. The goal is not to hide everything behind magic. We do this by generating a transparent `neptune.json` spec. You see exactly what is being built. It's a 'Glass Box,' not a Black Box. Our goal is to make infra work understandable, predictable, and automatable.
Right now it can infer things like databases, queues, API services, and background workers. It proposes a change to your schema. You approve it. Then it applies it. The AI doesn't have root access to your cloud; it has access to the spec. It is early and still rough. There are missing pieces and edge cases. We would love to hear where it breaks for you.
We would love any feedback, issues, ideas, and failures. You can test the beta (5mins) here: https://www.neptune.dev/try-it-now
Why we built this:
- Context Switching: Writing application code + Terraform + YAML breaks flow state.
- The 'Day 2' Wall: Most AI-generated apps are black boxes you can't scale. We wanted a transparent spec.
- IaC Fatigue: Managing state files for simple micro services feels like overkill.
If you want more context, we wrote a longer post here:
https://www.neptune.dev/blog/introducing-ai-platform-engineer