ALPHAGRID CONSOLE

Your personal agentic
development environment.

Built to work with your team, in real time.

free for personal use · runs on your machine · your code never leaves it
THE PROBLEM

The model was never the bottleneck. Knowing where to look was.

Run a few agents across a few projects and your day turns into tab-switching: which one is waiting, which one crashed, which plan still needs reading, which diff is safe. Grid replaces the terminals with one list of the decisions that are yours, and lets everything else keep working quietly.

claude · vertex
> implementing token bucket in middleware…
> editing api/keys.ts (3 of 6)
> running tests…
codex · helios
? Approve plan before touching storefront? (y/n)
  waiting 41m
claude · northwind
> closed 99 merge requests
✕ worker exited with code 1
> …
claude · cascade
> copying 12,408 objects
> rewriting srcset helper
> context 77%
claude · pilot
> make test 214/230
? which of the three billing models did finance pick?
→
NEEDS YOU · 3 3 things wait on you.
APPROVEAdd per-key rate limits to the public APIvertex4 steps · 6 files · 1 migration. Nothing runs until you approve.1h
REVIEWFeedback widget on all storefrontshelios2 files · +52 −0 · 3 builds green. Nothing is pushed yet.41m
DECIDEUsage-based billing for API keysvertexPer request, per key, or per plan?2d
18m◐ Close every MR untouched for a month · working 22m◐ Move catalog images to the new CDN · working 6m◐ Fix crash on resume when offline · verifying
EVERYTHING IS A CARD

A card says what you want, and shows what you need to know in seconds.

From a one-line bug fix to a masterplan with twelve subcards running in parallel, every piece of work is a card. It carries the plan, the questions, the session, the diff and the evidence, so you never have to reread a transcript to know where a piece of work stands.

1FEATUREvertexPREPARE
Add per-key rate limits to the public API 2Per-key quotas counted in Redis. Six files and one guarded migration, so existing keys keep working while it rolls out.
3THE PLAN · 4 STEPS 1Quota column on api_keys, behind your tap 2Token bucket per key in the middleware 3429 with Retry-After and quota headers 4Usage bar per key on the keys page
4● ONE QUESTION FOR YOUThe 41 keys in use have no limit today. Same 60 a minute on day one, or unlimited until you set one?A · 60/min, like new keysB · Unlimited for now
5● running locally · localhost:5173⌘1 · Opus · 1h
1Where it is, at a glance

The type, the project and the stage it is in, drawn the same way everywhere: on the board, in the office, on your phone.

2A summary that stays current

Two or three sentences, rewritten at every stage, so you know what the card does and where it stands without opening anything.

3The plan it will follow

Progress is counted against these steps, never guessed. A step added later shows up in Review as a change from the plan.

4Only the questions that are yours

Each with the options and what each one means, so a decision takes one pass instead of a conversation.

5The work, running

Its own worktree, its own ports, its session and its model, one click from the diff and the terminal.

WHAT A CARD CAN BEA one-line bug, an open question, a design change, a guard that never sleeps, or a plan with six subcards.
BUGpilotFix crash on resume when offlinemake test 214/230 · closes on greenVerify
RESEARCHheliosWhy have deployments become slower since the pipeline split?4 weeks of CI timings read · answer as a PDFDoing
MASTERPLANvertexUsage-based billing for API keys6 subcards · 3 running in parallelDoing
FEATUREvertex● needs youAdd per-key rate limits to the public APIplan ready · 1 question waits on youPrepare
FEATUREheliosTheme presets: 3 changes from Designmade from Theme presets v12 · diff attachedPrepare
RECURRINGnorthwindWeekly dependency bumpsevery Monday 06:00 · 14 packagesDoing
GUARDall projectsPipeline keeper: keep CI greenfound the card that broke quota.spec.tsDoing
BUGpilotFix crash on resume when offlinemake test 214/230 · closes on greenVerify
RESEARCHheliosWhy have deployments become slower since the pipeline split?4 weeks of CI timings read · answer as a PDFDoing
MASTERPLANvertexUsage-based billing for API keys6 subcards · 3 running in parallelDoing
FEATUREvertex● needs youAdd per-key rate limits to the public APIplan ready · 1 question waits on youPrepare
FEATUREheliosTheme presets: 3 changes from Designmade from Theme presets v12 · diff attachedPrepare
RECURRINGnorthwindWeekly dependency bumpsevery Monday 06:00 · 14 packagesDoing
GUARDall projectsPipeline keeper: keep CI greenfound the card that broke quota.spec.tsDoing
CHOREnorthwindClose every MR untouched for a month99 closed · a snapshot kept as the way backVerify
IDEAgridKeyboard-only board navigationbacklog · 2 of 4 checks · 1 question for designin the backlog
RESEARCHcascadeWhich image CDN is cheapest at our traffic?answered on the card: 41% cheaper, one caveatClosed
BUGhelios● needs youCheckout button hidden on iPad Safaribefore and after screenshots · needs your eyesReview
FEATUREnorthwindVX-412 · Invoice PDFs fit one pageimported from Jira · opening PR #214Deliver
MASTERPLANcascadeMove catalog images to the new CDN12,408 objects copied · 2 of 5 subcards doneDoing
CHOREpilotRaise test coverage above 80%Verify is the gate · 76% → 81%Verify
CHOREnorthwindClose every MR untouched for a month99 closed · a snapshot kept as the way backVerify
IDEAgridKeyboard-only board navigationbacklog · 2 of 4 checks · 1 question for designin the backlog
RESEARCHcascadeWhich image CDN is cheapest at our traffic?answered on the card: 41% cheaper, one caveatClosed
BUGhelios● needs youCheckout button hidden on iPad Safaribefore and after screenshots · needs your eyesReview
FEATUREnorthwindVX-412 · Invoice PDFs fit one pageimported from Jira · opening PR #214Deliver
MASTERPLANcascadeMove catalog images to the new CDN12,408 objects copied · 2 of 5 subcards doneDoing
CHOREpilotRaise test coverage above 80%Verify is the gate · 76% → 81%Verify
PLAN

Three ways a card begins, each one already refined.

The hardest part of working with agents is saying clearly what you want. Grid does that work with you: it asks the questions a senior developer would ask, sends each one to whoever can answer it, and turns the answers into a card an agent can finish without guessing.

Product writes the idea. Grid gets it ready.

Anyone can drop an idea in plain words, technical or not. Grid files it under a topic, checks what is missing, and asks. Pricing questions go to product, contract questions to finance, the Redis question to the backend developer. Each answer fills a readiness check. When all four are set, Grid proposes the type, the difficulty and whether it should become a masterplan.

Topics, filed for youFour readiness checksQuestions routed to the right personSketches and designs linkedImport from Jira and Azure DevOpsStart in Prepare, or Start building
Backlog · Monetise the API3 people on this idea
TOPICS
Monetise the API
vertex · 3
Search that never goes stale
northwind · 2
One-click onboarding
helios · 1
Field work offline
pilot · 2
NO TOPIC YET
✦ Grid files it for you
IDEAvertexby Sara · product● 3 questions open✓ ready
Usage-based billing for API keys
✓GOALTeam customers pay for what they call
○✓ACCEPTANCEmissingmetered per key, invoiced monthly
○✓SCOPEmissingexisting customers move on renewal
○✓TECHmissingreuse Redis, own DB index
✦ Per request, per key or per plan? → Sara · product
Sara Per key. Partners asked for it.
✦ Do existing Team customers move over? → Mark · finance
Mark On renewal, not mid-contract.
✦ Redis already runs sessions in prod. Reuse it for metering? → Jan · backend
Jan Yes, on its own DB index.
READY✦ Grid suggests MASTERPLAN · 4 subcards · HardStart building

Design it the way it should look. Grid writes the cards.

Chat with a model and watch the page change, on every screen size you support. Teammates click the design to comment. Every version is kept as a diff, so when you press Create cards, Grid compares the design with the running app and proposes one refined card per change, with the diff and the screens attached. The app never quietly drifts from the design again.

One design per projectAny model and effort4K down to a watchClick to commentHTML, PDF and videoImport a zip, a folder or the live appEvery change kept as a diffDiffs become cards
Design · vertex · keys.htmlv11v12 · 3 changes
Add a usage bar per key. Red above 90%.
✦ Updated keys.html: bar, threshold, tighter rows.
Sara Show it on the phone too?
✦ Done, 390 and 1440 match.
Describe a change…PSonnet · normal ▾
1440 · Desktop
API keys
prod-eu
prod-us
partner-acme
staging
API keys390 · Mobile
1Label in €?
Create cards from v12compared with the running appCreate 3 FEATUREUsage bar per keyFEATURERed above 90%CHORETighter key rows

Or just say it in one sentence.

Type what you want and Grid suggests the type, the difficulty, the project and the design it touches. Attach a screenshot, a log or a PDF. The card picks its own route: a bug reproduces before it fixes anything.

Type and difficulty suggestedProject guessed from your cardsScreenshots, logs, PDFsAccount, delivery and gates per card
New card⌘N
Checkout button is hidden on iPad Safari
TYPEBUG✦DIFFICULTYEasy✦PROJECThelios✦LINKEDCheckout design✦
ipad-checkout.pngattached
bug route · reproduce firstCreate
BUILD

Every card takes the route its type needs, with a fresh agent at each stop.

A feature walks all seven stages. A research card skips review and delivery because its answer is the deliverable. An agent card runs by itself; a guided card walks you through what only you can do. At every stop a new agent starts with a clean context and a contract to fill, so the card is the memory, not a chat that drifts.

ring: waits for you · dashed: skipped
InboxPrepareDoingReviewDeliverVerifyClosed
▣FEATURENew behaviour. You approve plan and diff.
✕BUGReproduced first, then fixed.
?RESEARCHA question. The answer is the output.
≣MASTERPLANA goal, split into parallel subcards.
↻CHORERoutine upkeep. Tests are the gate.
◈AGENTRuns on a schedule or an event.
☑GUIDEDOnly you can do it. Grid guides you.
A fresh agent at every stop

No 200k-token session wandering off course. Each stage agent reads the card and starts clean.

A contract to fill

A stage only ends when its sections are written: the handoff brief, the tasks, what changed, the evidence.

Only the stops it needs

Research never waits for a review, chores never wait for you. Routes are defaults you can edit per type.

The right model for each stop

Plan on a strong model, build routine work on a fast one. Set per agent and per difficulty.

Human approval, where you want it

By default you approve the plan and the diff. Set either to Automatic and the agent approves routine work itself, and asks you only when it touches a migration, a new dependency, security or production.

NeverAlwaysAutomatic
FOCUS

Forty cards in flight. Three of them light up.

Grey means an agent is working. Colour means you. Every page in Grid follows that rule, from Needs you to the board to the office, so the busiest day still has an obvious next thing to do.

●A plan to approve, a diff to review, a question to answer. Each arrives with the context to decide in one pass, never a bare “proceed?”.
●Also the things around the cards. A design with changes that have no cards yet, a pipeline a card broke, an environment that will not start.
●Everything else stays quiet. A muted timeline says who is working and how long ago they last did something.
EVERYTHING AROUND IT

The rest of the day, without leaving Grid.

The board is where the work moves. Around it sits everything a developer otherwise keeps in ten other windows, each one aware of which card it belongs to.

Needs you

APPROVE rate limits1h
REVIEW feedback widget41m
DECIDE usage billing2d
✓ approved · 2 left

Everything that waits on you across every project, oldest first. Nothing else asks for attention.

Design studio

v11v12

A design per project, on every screen size. Changes are kept as diffs and become cards.

Backlog and topics

Usage-based billing
goal · acceptance · scope · tech
3 questions open✓ ready to build

Ideas grouped by topic, with readiness checks that fill as people answer.

Agents and guards

Planner2 waiting
Builder3 running
Reviewer1 running
Pipeline keeperon it

Stage agents and guards you can read and edit, per stage and per card type.

Machine monitor

Rate limits · Doing
Feedback widget
vitest (left running)
load 3.4 · 12 cores$4.20 today

CPU, memory, run time and cost per session, and the processes agents left behind.

Run configs

api + workerstarting…running
web · :5173starting…running
postgres 16 · own dbseeding…running
logs · uptime 2h 14m↗ open

Start a whole stack on a card’s worktree like docker compose up, stop one service at a time.

Environment discovery

✓ docker-compose.yml
✓ package.json · 14 scripts
✓ .github/workflows/ci.yml
✓ Makefile · deploy
✦ 4 services · 2 pipelines found

Discover reads the repo and fills in how it runs, tests and deploys. Every value stays editable.

CI/CD tracker

buildtestdeploy
quota.spec.ts · 429 expectedred
Pipeline keeper found the cardfixing

Every pipeline per repository, the card that broke it, and a guard that fixes it.

Code on the worktree

api · wt/rate-limits ▾+ const bucket = tokenBucket(key, 60)+ if (!bucket.take()) return tooMany(res)  next()▍

Pick a card and every repository opens on its worktree, with the card beside the editor.

Terminals in tabs

⌘1 rate-limits
> pnpm test
> ✓ 41 passed
> ✓ quota.spec
> pnpm test
> ✓ 41 passed
> ✓ quota.spec
⌘3 feedback
> vite dev
> ready in 212ms
> GET / 200
> vite dev
> ready in 212ms
> GET / 200
shell · vertex
> git log --oneline
> e4216bb board
> 8e7af54 quota
> git log --oneline
> e4216bb board
> 8e7af54 quota
⌘4 old-mrs
> glab mr list
> !412 closed
> !418 closed
> glab mr list
> !412 closed
> !418 closed

One to nine terminals per tab, each on a card’s session or a free shell.

Library and artifacts

PDF Which ESP for invite wavesresearch
HTML Keys page · v12design
PNG ipad-checkout.pngattachment
MP4 Onboarding walkthroughdesign

Research reports, exported designs, screenshots and attachments per project, each linked to its card.

Accounts and secrets

admin-login-staging1Password
stripe-test-keyGrid vault
value••••••••••filled by Grid
never reached the model

Logins and keys stay in your password manager. Agents use them without ever seeing a value.

LOCAL FIRST

It runs on your machine, and your code never leaves it.

✓A local binary. Agents run where your repos are, with the provider you already pay for: Claude Code, Codex or GitHub Copilot.
✓A worktree per card. Your own checkout is never touched, and two cards never step on each other.
✓Secrets by reference. Logins stay in 1Password, Bitwarden or the Keychain; the value never passes through the model.
✓A way back from everything. Any card, change or conversation can be rewound to an earlier point.
TOGETHER

Work on the same board with your team, and from every device.

●Studio Macruns 6 agents ●Travel MacBookanswered 2 questions ○iPhoneapproves on the move Your teamlive on one board
Answer from any device while the work keeps running where it started. Share a workspace with a role per project, and see whose machine every card runs on.
PRICING

Free for personal use, with every core feature.

Pay when Grid starts connecting things: your own devices, then your team, then the whole organisation.

no device limitFree$0forever

Every area, unlimited projects, as many agents as your machine runs. No device limit. Commercial use for 60 days.

2 laptops and a phone, in syncSoloFreewhile Grid is in the alpha

Commercial use for one person. Sync across 3 devices and the phone app. Planned: premium plugins.

▦SRJBMKALTV
a team on one boardCrewPlannedwith the beta

Shared boards in real time, roles and approval limits, Jira, GitLab, Slack and Teams, audit log.

◎
a whole organisationFleetTalk to us

SAML and SCIM, locked policy, private plugin registry, self-hosted licence server.

Compare every plan →

Hand the work to the agents. Keep the decisions.

Grid Console is in alpha and free for personal use. Download it, point it at a folder of repositories, and the guided tour walks you through the real app.