Claude Code, at fleet scale

Run a fleet of agents.
Supervise one board.

Every Claude Code session on your machine, on one board that writes itself, under a memory the whole fleet reads. With loops and autonomy as first-class citizens.

Get early access See the whole machine

Runs on your machine, under your own Claude login. 14-day trial.

Self-writing kanban

Graph memory

Governed canon

Cross-agent review

Isolated worktrees

Multiplayer sessions

Alerts that reach you

Autonomy gate

Plain-English fleet status

Claude account auto-switch

orchestrate

Sessions3 of 95 of 9

Burst window holds at p99; writing the soak test now.6m
Buildfeat/rate-limit
Asking you: cap checkout retries at 3 or 5?11m
Buildfeat/retry-backoff
Warm pool is 5 of 5; handing members back to the queue.24m
Maintainchore/pool-warmup
Teardown clean across 200 reruns with the bounded retry.44m
Fixfix/teardown-retry
Resumed a 2 GB upload after a forced disconnect; ci green.31m
Buildfeat/upload-resume

6 more running4 more running

Inbox2 unread

blockedrun 118 · 09:41
the retry hides a real failure, not a flake
askingBuild · 09:53
Cap checkout retries at 3 or 5?
passedrun 119 · Review · 10:03
cleared by an agent that did not write the fix

pr 118 · fix/session-flakehead a41f2c9

session-runner.ts+2−1

88for (let i = 0; i < RETRY_MAX; i += 1) {

89 const err = await attempt(fn)

90 if (!err) return

91 await backoff(i)

92 retry()

92 if (err.name !== 'AssertionError') retry()

93 else throw err

Board23 itemssearch-keys · checks 4 of 6

identified12

in progress2

pr open2

ci1

merged6

gatedlive9 agents · 3.5M tokens today

Work faster in parallel.

Spawn fleets of sessions at once, track them all, including the ones you spawned in your own terminal.

the console

all projects6 running2 idle3 from your terminal1 needs you

running2m

Rate limiter drops bursts

apifeat/rate-limit

Rewriting the token bucket

running6m

Sessions leak on reconnect

apifix/session-leak

3 of 9 test files

needs you11m

Checkout needs a backoff

webfeat/retry-backoff

Cap the retries at 3, or 5?

running14m

Search misses item keys

webfeat/search-keys

Reading the index builder

running21m

Warm the pool before dispatch

infrachore/pool-warmup

5 of 6 members warm

running27m

Backfill the tenant column

apichore/tenant-backfill

Checking 14k rows

idle34m

Nav collapses on a phone

webfix/nav-collapse

Opened a pull request

running41m

Teardown fails intermittently

infrafix/teardown-retry

Reproduced it, 9 in 200

idle58m

Pin the docs toolchain

docschore/pin-toolchain

Waiting on review

The board writes itself.

Agents autonomously log, track, enrich, and execute tickets.

the board

board4 projects · 23 items

Identified12
no branch yet
- api2h
  Cart checkout times out
  from review 1284
- web5h
  Search should match order ids
  from you
- +10 more
In progress2
branch, no pr
- api4m
  Cap upstream retries at 3 with backoff
  feat/retry-backoff
  pool-3
- webstale 3d
  Index sessions by project for the picker
  feat/session-index
  pool-4
PR open2
pr open, no checks
- infra6m
  Resume interrupted uploads
  pr 1291 +312 -47
  pool-2
- api22m
  Bulk export runs off the queue
  pr 1290 +88 -12
  pool-5
CI1
checks running
- web11m
  Search matches keys as well as titles
  checks 4 of 6
  build, 2m left
Merged6
merged, branch gone
- api1h
  Flaky session-surface tests
  pr 118 to b4c7e21
- web3h
  Empty states name the next action
  pr 1286 to 8f21b04
- +4 more

Scroll to the merged lane

5 branches · 3 pull requests · 6 merges todayread 4s ago

Running the fleet is the easy part.

With built-in tools to solve the hard questions.

isolated worktrees

Will they repeat the correction I made yesterday?A canon every later run is judged against
What stops one drifting while I sleep?A gate, and three breakers it cannot widen
Do I still review every line myself?A second agent that opens by trying to refute
What did all of that use?Every token attributed to the run that spent it

Correct an agent without losing the run.

Steer the fleet without losing context.

the console, machine-wide

sessionfeat/retry-backofflive, 26m

agent pool-3

turns 7-8 of 41

Ranvitest checkout-retry11:42
Editedcap retries at the queue boundary11:45
Queued while running11:46ship the cap first, park the backoff
Ranvitest checkout-retry: 34 passed11:47
Turn 8
Repliedcap ships this turn; backoff parked11:47
Openedfollow-up: backoff curve11:48

enter queues into turn 8

Decisions outlive the session.

Corrections become rules. Every run after is judged against them.

the canon

canonretry-cap2 in · 1 rule out

what was claimed

we cap retries at 3, not 503-04 · you
5 retries on transient errors03-11 · pool-3

what the fleet follows

retries cap at 3 attemptsratified

They disagreed, so the gate asked you. 5 masks the flake.recalled by 5 runs since

commit messages state measured impactno question asked03-06 · you · nothing disputed it

Your rules stop being a file.

Disagreements get refereed by trust rules. Retract a bad session and its whole lineage leaves the canon.

the memory graph

sessionfix/flaky-authsessionfix/checkoutyou03-11

Ruleretries cap at 3 attemptsratified 03-11 · v2

run121run133pr118

edges written, not guessednothing reconstructed laterstale rules expire

Autonomy you can turn up.

Set how far the fleet goes alone. Three permissions never move.

the gate and its breakers

gateautonomy3 of 10 always held

gatedfull

an unrecognised value resolves to gated

Every row is something the fleet may attempt. The two columns are the verdict at each autonomy setting.
what the fleet may do	gated	full
deliberate among themselves	automatic	automatic
spec, propose and enrich the work	automatic	automatic
open a work item and a branch	automatic	automatic
push a branch, open a pull request	automatic	automatic
dispatch an agent into a worktree	recorded, not run	automatic
merge a pull request that is green	waits for you	automatic
run the loop overnight, unattended	waits for you	automatic
no setting reaches below	gated	full
widen its own autonomy setting	refused	refused
resolve a review thread it wrote	refused	refused
keep merging without a review	halts the loop	halts the loop

Automated, auditable code reviews.

Run reviews pre-PR. Audit your most common mistakes. Build the rails that heal themselves.

the review

review14 days80% refuted

reviews ran38pre-PR, per push

findings filed214all of them tried

died refuted171the noise, named

promoted4341 fixed and cleared

died refutedpromoted

reviewpr 1185 threads · 4 refuted · 1 confirmed

raised0 of 5 reached you

by agent/pool-2, judged by agent/pool-4

majorcorrected from criticalconfirmedsession-surface.test.ts:212

the retry hides a real failure, not a flake

Triggerrun 47 of 200 fails the assert, not the timeout

expect(rows).toHaveLength(3)The other four named no trigger.

gate majorhead a41f2c9

The record has no delete.

Every action logs its actor, its trigger, and what cleared it.

the trail and the canon

trailpr 11834 events · oldest first

session-flakemoved to in progress09:02:41work.startedhook/post-checkout
run 118reproduced the flake in 9 of 200 runs09:07:18run.recordedagent/pool-2+4m37s
retry-capretry a flaky test up to 3 times09:14:02canon.ratifiedagent/gate+6m44s
pr 118Flaky session-surface tests09:38:20work.pr_openedhook/pre-push+24m18s
run 118the retry hides a real failure, not a flake09:41:55review.blockedagent/pool-4+3m35s
run 119cleared by an agent that did not write the fix10:03:12review.passedagent/pool-4+21m17s
retry-capretry only on transient IO, never an assertion10:26:30canon.ratifiedyou+23m18s
retry-capthe earlier rule is kept, not edited10:26:30atom.supersededengine+0s

Ten kinds, a closed vocabulary the engine emits. No row is free text.

Every token names its work.

Token usage attributed to the run, branch, and model that spent it.

what moves the price

usage19 runs · 3.5M tokens today

session-flake merged 10:26443k

sonnetReproduced the flake, 9 of 200 runs64%284k
opusWrote the fix, 3 files22%96k
opusVerified it independently9%41k
haikuBoard, trail and canon writes5%22k

today, all projects4 projects · 1 held

api9 runs1.9M
chore/autonomy0 runs0held by the gate
web6 runs1.1M
infra3 runs480k
docs1 run20k
today19 runs3.5M

Counted on your own Claude plan. No API key, and no usage bill from us.

Nothing is addressed to us.

Runs on your machine, on your own Claude subscription. No API key.

local first

graphsession-flake6 nodes · 0 to us

your machine

what leaves it

model callto your Claude account
git pushto your own remote
to us0 byteson any network, at any time

12+fleet systems: board, memory, gate, review, spend
0bytes: of your code reach us
NoAPI key: it runs on your Claude plan
$39a month: flat, whatever you run

Spawn your first agent team.

Install it, point it at a repository, and dispatch your first five.

Get early access

$39 a month, 14-day trial. Cancel from the app.

Run a fleet of agents.Supervise one board.