hritvikgupta/nimbus

Nimbus is an AI agent that lets you manage your cloud infrastructure by talking to it in plain language, instead of switching between separate cloud consoles, terminals, CI dashboards, and infrastructure files. It reads your code, understands your architecture, connects to your real AWS and GCP accounts, and can fix problems in your repositories by opening pull requests. The project is built around a few core ideas. You describe the outcome you want, Nimbus proposes a plan, and you approve it before anything runs. Reading information from your cloud accounts happens automatically, but any action that changes a billable resource always waits for your explicit approval first. Everything the agent says is meant to be backed by something it actually read or ran, such as a specific file, a log line, or a live resource, so answers are grounded rather than guessed. Teams share one workspace per project, including channels, cloud connections, and an architecture canvas, while private chats with the agent stay personal. Key features include a chat interface built on a ReAct style tool use loop, a live architecture canvas that draws system diagrams such as load balancer to app to database as you design them, and connections to AWS and GCP through per user MCP servers using your own credentials. It can also connect to a laptop or a CI runner you control, then drive a local Claude Code session step by step to fix a repository, before pushing a branch and opening a pull request. Credentials are encrypted at rest, never shown or logged, and every action leaves a server side audit trail. Under the hood it is a React single page app talking to a thin Express API server, backed by SQLite for storage and Zod for validation. The AI layer uses the Vercel AI SDK with Databricks as the main model provider and OpenRouter as a fallback, plus the Model Context Protocol for connecting to cloud services. The easiest way to run it yourself is with Docker Compose: clone the repo, copy the example environment file, generate an encryption key, fill in your AI provider details, and start the container. The project is released under the Business Source License 1.1, which is source available but places restrictions on competing commercial use.