R·C·C agenticframework.in →
Remote Command Console

Every agent session.
Every machine.
One console.

RCC is the ops layer above your AI coding agents. Claude Code, Codex, Cursor, and Hermes sessions across all your machines on one live board — watch them work, take any of them over, approve tool calls from your phone, and let scheduled runs keep working while you sleep.

Self-hosted. No RCC cloud your agents run through — it installs inside your own environment, your machines stay yours, and you bring the AI subscriptions or API keys you already have.

the fleet, live
api-refactorvps-01 · ⚡
nightly-testsdev-box · ⚿
⚠ Bash  nginx -s reload
game-playtestvps-01 · ✍

What it does

The whole fleet on one board

Every session on every machine, sorted into live lanes: working, waiting on you, idle. Board, project, grid and map views — plus a PWA on your phone.

Adopt and drive any session

A session someone started in a terminal this morning? Take remote control of it from the browser or Telegram and keep the conversation going.

Approvals from anywhere

Tool calls wait for your allow or deny — in the console, or as Telegram buttons in your pocket. Hit “always” once and that command stops asking.

Scheduled runs

Cron-spawned sessions on each machine's own clock: nightly test triage, morning summaries — they run whether the console is open or not.

A watchdog and a morning digest

A session that goes quiet mid-run pings your phone. Each morning: one message with what ran overnight and what's blocked on you.

Ask the fleet

Question your entire transcript history — “which session fixed the nginx config, and how?” — answered with citations to the exact sessions.

Squads — give it a mission, not a message

This is where RCC stopped being just a remote command console and started becoming a remote control console. Give it an engineering objective, assign roles to different agent sessions — even across different harnesses and machines — and let them work the mission together: Claude as developer, Codex as independent reviewer, Claude again for remediation, testing and operating the fix.

Structured handoffs

Findings from review route straight back to remediation; the reviewer re-verifies the actual fix, not a summary of it.

Evidence, not claims

Tests and other checks are captured by RCC itself — a stage doesn't move on because an agent says “done.”

Session reuse

The same adopted sessions carry the mission forward stage to stage, instead of spinning up fresh, contextless agents each step.

You set the boundaries

What agents are allowed to do, how autonomous a mission runs, and exactly which situations still need you — you decide, always.

Routine work — implement, review, fix, verify — runs end to end without you standing in the middle as the message router between agents.

How it holds up

Agents dial outNo inbound ports on your machines — each agent opens one outbound connection to the hub.
Private by defaultThe hub binds localhost. Public access is a token gate behind TLS, killed with one command.
Hash-chained auditEvery approval, denial and dispatched command is logged tamper-evident and exports to CSV.
Scoped tokensPer-user tokens with view/control scopes and per-project rights — a reviewer can't touch prod.
Bring your own agentsNo forced provider — Claude Code, Codex, Cursor and Hermes, on the subscriptions or keys you already pay for.

Not a download — a trial

RCC runs my own fleet every day; that's the demo above and the video — Squads included, tested first on a disposable project and now running on a real build. It isn't packaged for self-serve yet, on purpose: the first few people who need this get private trial builds, onboarded personally, in exchange for honest feedback about what you'd run it on.

One line about your setup is enough — how many machines, which agents.

tell me what you'd run it on →