Mission control for a fleet of coding agents.

Every session, subagent, MCP server, and background bash shows up as a node, and it moves as they do. There's no refresh button to hit.

Every permission request from every agent lands in one queue, sorted by risk. Approve or deny them in batches, or write a rule once and let the routine ones clear themselves.

Nodum blocks the commands you never want run (rm -rf /, fork bombs, raw disk writes) and holds the merely risky ones for you to check first. Worth having when an agent is working unattended.

Kill a runaway agent on the spot. Put a token or dollar budget on each one. Let Autopilot wave through the safe actions, or roll back what an agent did with a single click. All of it gets written to the audit log.

Codex CLI sessions show up on the same graph as your Claude Code ones, transcripts and all. Toggle them on or off, or narrow the view to whatever's run recently.

A record of every approval, guardrail hit, budget stop, and rollback. Search it, filter it down, and export whatever's left.

Hand one task to a set of repos and Nodum spins up an agent for each. You watch them all on the graph, and their approvals come back to the same inbox.

Save an agent profile once (its instructions, model, and tool policy) and reuse it. Spawn or dispatch a fleet from it, or export it to Claude Code's own ~/.claude/agents folder.

A running feed of what's happening across every agent: fleets going out, subagents starting, approvals decided, guardrails firing, sessions wrapping up.

Click a node to read its full history, from the messages to the tool calls, token counts, and roughly what it cost. Even a 5,000-line transcript scrolls smoothly.

Start a new session, managed (SDK) or interactive (PTY), without leaving the app. Pick an old one back up with one click. It all sits in color-coded tabs.

Watch every Read, Edit, and Write hit your project, with a git diff and preview for each file a session touches.