AgentHelmFleet operations console for coding agents
Every agent. One helm.
You run a fleet now — Claude Code on the refactor, Codex on the tests, a shell agent on the migration. AgentHelm puts every session's status, spend, and pending decisions on one board.
Runs locally · Nothing leaves your machine · No account to start
The actual console, running its demo fleet: four agents, two decisions waiting, 1,520,870 tokens burned — at a glance.
Reads the session stores of Claude Code · Codex · GitHub Copilot · OpenCode · Cursor-class agents
Fleets don't break down in the coding.
They break down in your attention.
The agents are fast. The bottleneck is the human switching between terminals, catching stalls late, and hand-carrying context from one session to the next.
Prompts you never see
A permission request sits in a background terminal for an hour. The agent isn't slow — it's waiting on you, and nobody told you.
Stalls you notice late
An agent finished — or died — forty minutes ago. You find out when you go looking, not when it happens. Stall detection should be automatic.
Handoffs that lose the thread
A session ends and its context ends with it. The successor starts cold — same task, repeated mistakes, decisions re-litigated.
Approvals inbox
Every decision your fleet needs. One queue.
Permission prompts, questions, and decision requests from every tool — sorted, batch-clearable, logged.
Two real items from the demo board — both answerable without opening a terminal.
- One queue across every tool — Claude Code, Codex, Copilot, OpenCode, Cursor-class
- Keyboard-driven — j/k to move, a approve, d deny
- Nothing waits silently — what's blocked on you, and for how long, is always on the board
Burn accounting
Know what today's fleet cost. Before the invoice does.
Token and cost burn per agent, rolled up per project — read from the session logs your agents already write.
The demo fleet, exactly as the board reports it. Costs are estimates from session logs — and labeled as estimates.
- Per agent and per project — see exactly which session is spending
- Computed from session logs — never self-reported by the agents
- The total stays in view — the board header carries the fleet burn at all times
Handoff briefs
Continuity is a document, not a memory.
When an agent finishes or gets replaced, AgentHelm writes the brief — so the successor picks up the task, not the archaeology.
A brief the product generated from the demo board's own data.
The whole helm, in one console.
One board, whole fleet
Every agent session across every project: working, waiting, idle, done, stalled — with current task and last activity. Stall detection is automatic, not something you notice by luck.
Approvals inbox
Every permission prompt, question, and decision request from every tool, in one queue. Clear decisions in a batch instead of babysitting terminals.
Burn accounting
Token and cost burn per agent, rolled up per project — computed from your agents' own session logs, with estimates labeled as estimates.
Handoff briefs, automatic
When an agent finishes or gets replaced, AgentHelm writes the brief — task, decisions, files touched, open items — and hands it to the successor.
Cross-tool, Cursor included
Claude Code, Codex CLI, GitHub Copilot, OpenCode — and the Cursor-class agents. Adapters read each tool's own session store; a generic adapter tracks any long-running agent process.
Local-first and instant
The console runs on your machine and reads local session data. No account to start, nothing leaves the box, interactions stay under 100ms, and the whole board is keyboard-driven.
Running in three steps.
01
Start your agents
Claude Code on the refactor, Codex on the tests — exactly as you do today. AgentHelm changes nothing about how your agents run.
02
AgentHelm finds them
Adapters read each tool's own session store. Every session appears on the board with its status, task, and burn. Zero config.
03
You steer from one board
Decisions land in one inbox. Handoffs write themselves. The fleet stops depending on which terminal you happen to be watching.
Flat plans from $0.
Free for up to 3 agents. Pro is $19/mo with the approvals inbox and burn accounting. Fleet is $49/mo for working fleets. Published prices, no sales call — the way fleet tools should be sold.
Take the helm.
Your fleet is already running. Give it an instrument panel — start free with up to 3 agents.