See every AI worker you run.
Every organisation running AI workers hits the same wall: each machine only knows itself, and finding out what your fleet is actually doing means reading logs on one box at a time. Fleet View is the first tool that gives you that per machine, out of the box — from the beacon every machine already publishes once it runs a session on a board.
Four questions, one screen: what is each machine doing, is it stuck, what is slowing it down, and what else could it be doing.
Updated October 30, 2026 · v0.16.12 · ✅ Supported · part of the 321-tool board · release notes
What is each machine doing right now?
fleet_status reads every machine's beacon — a small JSON file each
box rewrites every 60 seconds and the pad's own git sync already carries — and folds
them into one summary line plus a card per machine: lanes busy/total, one line per
held lane (ticket, state, how long, model, spend), and today's closed count. No
separate service, no extra agent to run.
Is it stuck?
A machine holding 16 lanes looks identical from the outside to one that died holding 16 claims — until the stuck detector tells them apart. Silent (stopped publishing while still holding lanes), all-stalled (alive but nothing is moving), refill-dead (writing but never dispatching), crash-loop, and throttled-out each get their own reason and their own red banner, never a generic "something's wrong."
What is slowing it down?
Every machine and session gets a limiter verdict from MEASURED signals — ledger slot states, CPU/GPU samples, throttle signals — never a guess: llm-throttled, cpu-bound, gpu-bound, io-bound, orchestrator, human, or the healthy llm-bound. A throttled machine's badge carries its own reset countdown, so "wait" and "broken" never look the same.
What else could it be doing?
spare_capacity turns idle CPU/GPU/LLM headroom into a ranked
suggestion with the exact call to run it: a throttled or LLM-bound machine with idle
cores gets pointed at compute-only work — full test suites, fuzzers, benchmark
reruns — from your project's own spareWork jobs; a CPU-bound machine
with LLM budget left gets offered cheap-model lanes instead. Nothing runs on its own
unless you turn autoRun on.
One call per machine.
The beacon is on by default the moment a machine runs a session on a board — nothing to configure for a single box. Running more than one?
Run fleet_setup on each machine
Makes that box board-ready in one idempotent call — identity, code location,
pad hygiene, compute profile — and it shows up in the Fleet tab on its very next
beacon write. The setup guide has this as a one-click, optional step: "More than
one machine? Run fleet_setup on each; the Fleet tab lights up."
fleet_setupOpen the Fleet tab
The board's Fleet tab renders exactly this data as one card per machine,
auto-refreshing every 30 seconds — the same read fleet_status
answers in chat.
get_boardOpt in to the cloud relay
Check the fleet from your phone without opening a terminal: an opt-in relay to your featureboard.ai dashboard, off by default. Want to file work and answer blocked lanes too? Run every AI worker from one screen with the Control Plane.
Stop guessing what your fleet is doing.
Free for personal and public work — the beacon is already running the moment you open a second board on a second machine.