The value isn't a dozen bundled apps — it's cross-domain execution without a person becoming the integration layer. A signal in one system becomes one governed mission that uses capability and context from the others: one identity, one evidence trail, one auditable history. The stack is six governed domain agents built from real open-source cores, composed as suites and connected by rAIls. This page stands the whole thing up.
The suite ships as one Compose stack in deploy/apps of the public agentic-os repo (app images are public on GHCR — no login). "One-click" honestly means one command — a single docker compose up on a host, or one helm install in a cluster. Either way a deployment is itself a governed mission — planned, gated on a human approval, backed by evidence, and unwound by a saga if it fails.
You need Docker (with the Compose plugin) and an OpenAI-compatible model endpoint (a local one is fine).
# clone the stack + run one command git clone https://github.com/redevops-io/agentic-os cd agentic-os/deploy/apps cp .env.example .env # set a model endpoint + secrets docker compose up -d # the full suite, images pulled from public GHCR
Then open the operating console:
Console / cockpit http://localhost:8080/ # default HTTP_PORT=8080 in deploy/apps/.env Console health http://localhost:8080/health
Out of the box SHADOW_MODE=true: agents draft actions and stage them for approval in the console; nothing is sent to a real CRM, social account or customer until you approve.
Run the governed runtime in your cluster via the Helm chart, with a hardware/cloud profile. Because every model endpoint is OpenAI-compatible, the same chart serves any target by pointing at the right endpoint — no per-vendor adapter.
git clone https://github.com/redevops-io/agentic-os
helm install rr ./agentic-os/deploy/reference-system/redevops-runtime \
-f ./agentic-os/deploy/reference-system/redevops-runtime/profiles/nvidia-kubernetes.yaml # or amd-kubernetes · nvidia-nim · …
Profiles live in redevops-runtime/profiles: nvidia-kubernetes, nvidia-nim, amd-kubernetes, amd-instinct, nemo-retriever. The post-install gate runs acceptance checks and fails the release if they don't pass — the deploy is a mission.
Each cloud target runs as a governed, approval-gated deployment mission (preflight → plan → approve → provision → verify → drift) via the matching skill — you tell Sidekick the target, it picks the runbook. See the matrix in §2.
# with the Sidekick agent up (§3), just say where:
❯ Deploy the rAIlsApps to AWS EKS (staging). Stop at the approval gate.
❯ Provision a Proxmox k3s node and deploy the stack to it.
--flagship.Safe by default. With the default SHADOW_MODE=true, consequential actions are staged as gated missions that park on a human approval in the cockpit before anything changes.
Deploy on a suite's native cloud where it has a first-party home, or on-prem (k3s) otherwise. The deployment-* skills run each target as a governed mission; the Helm profiles serve the model endpoint.
| Suite | Native cloud | On-prem / self-host |
|---|---|---|
| AWS | EKS + Bedrock · deployment-aws | k3s |
| Azure | AKS + AI Foundry · deployment-azure | k3s |
GKE + Vertex · deployment-gcp | k3s | |
| NVIDIA (NeMo / NIM) | hosted NIM · nvidia-nim, nvidia-kubernetes | GPU k8s · same profiles |
| AMD (Instinct / ROCm) | — | amd-kubernetes, amd-instinct |
| DigitalOcean | DOKS · deployment-digitalocean | k3s |
| CrewAI · LlamaIndex · PydanticAI | framework — runs beneath the runtime | compose / k3s |
| proxmox (self-host default) | — | deployment-proxmox · make proxmox-agentic-os |
Skills live in sidekick-devops/skills. Each is the same governed mission with a different arm: preflight → plan → approval gate → provision → configure → verify → drift.
The agent is Sidekick (the Mission Runtime) from deploy/sidekick-devops. It runs governed missions and gates changes on approval; the guide app can answer usage questions over the apps' own docs and capabilities. It needs one thing you choose — an OpenAI-compatible model endpoint.
git clone https://github.com/redevops-io/agentic-os # if you have not already cd agentic-os/deploy/sidekick-devops cp .env.example .env # set MODEL_ENDPOINT to your OpenAI-compatible endpoint (local vLLM/llama.cpp or a hosted API) MODEL_ENDPOINT=http://host.docker.internal:8000/v1 docker compose up -d # Sidekick runs as its own compose stack
Once it's up, drive a cross-system mission in plain language — a goal in one system becomes one governed mission spanning the others:
A client's position breached its 12% concentration limit — what should we do?
The agent opens one mission: market intel confirms the move → portfolio sizes the breach → tax prices the trade against the budget → compliance checks restrictions → the trade exceeds advisor authority, so it parks on an approval gate. Unauthorized actions are blocked at the gate, and the recorded trace supports audit replay.
How do I connect our billing data, and which app owns dunning?
The guide answers from the stack's own capabilities: billing wraps Lago, dunning is a billing capability, and here's the exact config — a usage question answered by the same agent that runs the work.
Set ALERT_WEBHOOK_URL and each approval gate pings a human for sign-off with a cockpit deep-link — the fast path from a detected issue to an approved fix. Gated actions do not execute without that sign-off.
Questions, or stuck on a step? Ask in #agentic-apps on the ReDevOps Slack, or email info@redevops.io — the team and community help with deploys, connectors, sources and governance there.