Infrastructure an AI agent already knows how to use.
CLI-first. SSH-based. JSON by default. Every rdc command is built for autonomous agents: no GUI, no browser automation, no custom integrations.
AI agents are powerful, but most infrastructure fights them
An AI coding assistant can write, test, and debug code autonomously. But when it needs to deploy, check status, or manage infrastructure, it hits a wall: GUI-only dashboards, OAuth browser flows, unstable APIs, and responses that blow the context window.
See the timeline
Three steps to agent-managed infrastructure
Configure
Drop an AGENTS.md template into the project root. The AI assistant reads it on startup and knows every rdc command.
Discover
Connect the assistant to rdc mcp serve and it gets typed tools for every operation. Or run rdc --help-all for the full command reference.
Operate
Deploy, monitor, rollback. All through CLI commands with JSON output. No GUI, no browser, no custom integration.
Why rdc is built for AI agents
Every design decision in Rediacc prioritizes machine-readable interfaces. Non-TTY detection auto-switches to JSON. The consistent envelope format means agents parse every response the same way. SSH-based auth means no OAuth flows or token management.
Why rdc is built for AI agents
| Traditional platforms | Rediacc (rdc CLI) |
|---|---|
| GUI dashboards: requires browser automation or screenshots | CLI-first: every operation is a shell command |
| OAuth/browser login: agents can't authenticate | SSH keys: agents authenticate like any CLI tool |
| HTML/unstructured output: 4x token usage, fragile parsing | JSON envelope: consistent schema, ~200 tokens per response |
| No discovery: agents need docs or hardcoded API calls | MCP server: typed tools discovered at runtime |
| Cluster changes: web consoles, wizards, multi-step runbooks | rdc cluster fork: one command, one JSON reply, whole cluster |
Built for autonomous infrastructure management
Nothing to integrate
The agent reads one file in the project root and knows the whole command surface. There is no SDK to install, no schema to hand-write and no credentials to broker.
Answers stay small
Replies come back structured, so an agent parses them the same way every time and spends its context on the task instead of on scraping a dashboard.
Destructive work asks first
Anything that removes something previews it before acting, and automation has to opt in explicitly. An agent can move quickly without one bad run being unrecoverable.
See related solutions
Let the AI agent manage the infrastructure
Install the CLI. Add the AGENTS.md template. The AI assistant handles the rest.
Start free trial