Atmos is the open-source runtime that builds, authenticates, and ships Terraform, Kubernetes, and containers — the same way everywhere. Auth, secrets, vendoring, and CI are built in. Stop stringing together 25 tools.
Frontend teams move fast because the framework is consistent and predictable. Atmos brings that to infrastructure. Everything is declarative — the same stack configuration, the same commands, everywhere. Skills teach agents your domain, and every command documents itself. So agents don't string together 25 tools and pray — they operate one provable, end-to-end framework.
The tools, workflows, dependencies, and validation — everything you used to script together — is declared and wired end to end. Agents drive one complete system, not a pile of glue scripts.
22 portable skills in the open Agent Skills format hand agents exactly what they need about your stacks, components, and workflows — working across Claude Code, Cursor, Gemini, and Copilot. Publish your own.
Atmos exposes itself over the Model Context Protocol, so any MCP client — Claude Code, Cursor, VS Code — can query and drive your infrastructure as native tools. No custom integration.
Three steps from config to authenticated infrastructure workflows.
Stacks, components, identities, secrets, and stores live in one declarative graph.
Atmos identities feed Terraform, stores, emulators, and CI without bespoke wrapper scripts.
The same commands build, authenticate, and ship on your laptop and in CI, identically.
Point every environment at the same reusable Terraform root modules and treat the rest as configuration — eliminating code duplication, custom bash scripts, and complicated tooling with one tool to rule them all.
And it isn't just environments — auth, secrets, workflows, and vendored dependencies are all declared the same way, so the entire platform stays readable, reviewable, and in sync.
Auth, secrets, vendoring, caching, toolchain, workflows, CI, and AI are part of the runtime — not a pile of plugins you wire together.
Plan and apply across every component in dependency order, with bounded concurrency. Backends and providers are generated for you. Drift is caught automatically.
Atmos lists the Terraform instances and resolved stack values it will pass to Terraform.
Helmfile is a first-class workload. Atmos models Kubernetes releases beside the rest of your stack, with the same CLI you already use for Terraform.
Atmos starts a local k3s sandbox off camera, installs declared Helm tooling, then deploys the release from the stack.
Containers and dev containers are workloads too. Spin up cloud emulators locally so your whole stack runs on your laptop — no account required to iterate.
Develop against a local emulated cloud, then ship the identical config to prod.
Same command, same auth, same secrets, same toolchain — whether you run it locally or in a pipeline. And Atmos is git-aware: it detects what changed and plans or applies only the affected components, so CI does exactly the work that changed — nothing more.
What works on your machine works in CI — without any additional GitHub Actions or messy bash scripts, because it is literally the same runtime.
Declare required secrets next to a component, initialize or rotate them through Atmos, and inject them into commands only when the component runs.
Atmos initializes declared secrets, lists their status, rotates a value, and injects it into a component command without printing the secret.
Forget a flag and Atmos asks which stack and which component you meant. Every command follows the same verb-noun grammar, so the CLI stays discoverable as your platform grows.
A consistent, discoverable CLI that guides you instead of fighting you.
Everything is pluggable — and your agents can use it the moment it exists. Atmos ships a catalog of agent skills and an MCP server, so any agent can install what it needs and drive your infrastructure directly.
Free and open source. Run your first stack in minutes.
