● spec-124 and spec-125 both depend on how the log endpoint pages through deliveries.
Pagination
PaginationHow should GET /webhooks/deliveries paginate?Enter to select · ↑/↓ to navigate · Esc to cancel
How should GET /webhooks/deliveries paginate?
Your coding agents, working like an engineering team.
Work gets a spec before it runs and acceptance criteria on every card, and a second agent can check it before it closes. You keep the CLIs you already pay for.
Founding price $15/month $19
For Windows and Linux.
Bring the subscriptions your CLIs already use.
dcoder
● 2 agents need you
Search agents, branches… Ctrl K
● Reading spec-123 and the current signer.
● Read(src/webhooks/sign.ts)
⎿ Read 52 lines
● Signing with HMAC-SHA256 over the raw body plus the timestamp, per the spec.
● Update(src/webhooks/sign.ts)
⎿ Updated src/webhooks/sign.ts with 24 additions and 3 removals
● Write(src/webhooks/sign.test.ts)
⎿ Wrote 58 lines to src/webhooks/sign.test.ts
● Bash(pnpm test webhooks/sign)
⎿ ✓ 9 passed (412ms)
✢ Writing… (3m 12s · ↓ 4.1k tokens)
● Reading spec-125 first.
● Read(web/src/pages/webhooks/index.tsx)
⎿ Read 74 lines
● Write(web/src/pages/webhooks/history.tsx)
⎿ Wrote 128 lines to web/src/pages/webhooks/history.tsx
● The history screen needs the index on (created_at, id) to page fast. Adding the migration.
● Write(migrations/0042_delivery_history_index.sql)
⎿ Wrote 6 lines to migrations/0042_delivery_history_index.sql
Bash command
pnpm db:migrate --env testApply 0042_delivery_history_index to the test databaseDo you want to proceed?Run pnpm db:migrate --env test?
› Why did p95 on POST /charges go up this week?
• Explored
└ Read src/routes/charges.ts, src/idempotency.ts
Search idempotency_key in migrations
• Ran pnpm db:explain charges:create
└ Seq Scan on idempotency_keys (rows=1.8M, 212 ms)
• Migration 0039 rebuilt idempotency_keys and dropped the unique index on (account_id, key). Every POST /charges now scans the table: p95 went from 41 ms to 230 ms.
Fix: restore the index concurrently. I wrote it as migrations/0043_idempotency_index.sql, left unapplied.
Worked for 4m 12s
gpt-5.3-codex high · 61% context left
The top bar switches between three views of the same work.
Thread, Grid and Kanban all look at one workspace. Switching views never stops an agent: the terminals keep running behind whichever view is open.
The same process dcoder-terminal is built with.
Agents write code faster than anyone can read it, so the checking has to be part of how the work runs. Here it is: a spec before the code, the right model for each task, parallel work with boundaries, and a reviewer that did not write the code on the cards that name one.
- 01 Specify with the pilot You talk the work through and the pilot writes the PRD, then cuts it into specs with a plan, acceptance criteria and dependencies. Nothing leaves the Backlog until you promote it.
- 02 Pick the actor for each card A card names who runs it: CLI, model, effort, permission mode, skills and MCP servers. The expensive model can plan and review while a cheaper one builds a slice that is already specified.
- 03 Run in parallel, inside boundaries Cards whose dependencies are met run at the same time, each in its own pane, up to the slots you allow. Work that runs side by side gets separate files to touch, written into each card.
- 04 Review before it closes A card can carry a check command that runs when the worker finishes, and a reviewer that checks the delivery criterion by criterion. A rejection goes back to the Queue with the reason written on the card.
- 05 You decide An agent that needs an answer stops and the pane turns to Needs you. You promote cards, answer questions and keep branches and merges in your own hands.
Loading the interactive demo
The same sample project, opened on the Kanban. Open a card to read its spec and the reviewer's notes.
It runs the agent CLIs already installed on your machine.
- Claude Code
- opencode
- Codex
- Antigravity
- dcoder-agent
- Kimi
- Grok
- Copilot CLI
- deepclaude
- mimo
- Cursor Agent
- Qwen
- MiniMax
- any shell
The brighter ones are first class: the board knows when one finishes or stops to ask you something, and most of them report their cost live. Any other CLI runs from a command you write once.
pilotworking
Pilot · Opus
Flaky retry test on CI
Fleet cost$0.00
retry-reproWorker · Sonnetnot open
spec-130Queue
src/webhooks/retry.ts, src/webhooks/retry.test.ts
ci-traceWorker · Sonnetnot open
spec-131Queue
.github/workflows, read only
The pilot splits the work across a crew, and each member carries a card.
When a piece of work can be divided, the pilot opens crew members in new panes. Each one starts from a written mission and a card, and everything it reports lands on that card, so the history explains itself to whoever comes next.
- Every crew member gets its mission in writing, with the part of the code that is its own.
- They work at the same time, each in its own pane, and every report comes back to the pilot on its own.
- The pilot reads each report and the member's screen, sends a correction when the work falls short, and closes the card when it is done.
A crew member never moves its own card. The lane belongs to the pilot and to you.
Everything you would leave the window for is already in it.
Switching apps while agents are running is how you lose track of what they are doing. These come with dcoder-terminal so you can stay where you are.
Git
One Git window for every repository in the project, where you stage by hunk, resolve conflicts in a merge editor, and open and merge pull requests on GitHub, GitLab or Azure DevOps.
Editor and files
Monaco with a file tree and project wide search, and a selection you can send to the agent in focus.
Project memory
dcoder-memory keeps what the agents learn about the project in pages they all read and write, and the Memory window lets you browse, edit and map them.
Skills shelf
Drag a skill onto a pane and the instruction to read it is typed there, ready for you to send.
Voice
Dictate into any pane from a global hotkey, or switch on conversation mode to talk to the app in real time, and it asks you out loud before anything destructive.
Remote access
Pair your phone with a QR code to follow your agents and answer the ones waiting on you, end to end encrypted, and only once you turn it on.
spec-123 HMAC signature main · Worker · Sonnet 96.4k · $1.12 14m 13s
● Bash(pnpm test webhooks/sign)
⎿ ✓ 9 passed (412ms)
✻ Writing… (3m 12s · ↓ 4.1k tokens)
spec-121
Delivery table and migration
$0.9200 real ·
$2.31 API ↑184k
↓21k tok
$2.31 API ↑184k
↓21k tok
ledger-api workspace
$21.84 real · $63.20 API
↑4.8M ↓402k tok
You see what each agent costs, in dollars, while it runs.
The seat is the only thing you buy from us. Tokens go from your CLI straight to the provider you already use, and the board, the specs and your code stay on your disk.
- Cost in the pane header, on each card and summed across a PRD, on the CLIs that report it.
- A dashboard of tokens and cost by CLI, model and project, with subscription usage priced at API rates and kept apart from real spend.
- The licence check is the only thing that talks to us, and your code and prompts are never sent our way.
- Remote access stays off until you pair a device, and its traffic is end to end encrypted.
One plan, priced per seat instead of per token.
You pay for the app, and the models stay on your own accounts. Opening another terminal adds nothing to this line of the bill.
Founding
Founding priceEverything in the app, for whoever subscribes early.
$15$19per month
Locked for as long as you stay subscribed, so it never rises to $19 under you.
Everything, from day one:
- Real terminals in a grid, for any interactive CLI
- First class integration with the main agent CLIs
- A board in Markdown, inside your repository
- PRDs, specs and the pilot, in the Thread or in a terminal
- A scheduler with as many slots as you allow
- A check command and an AI reviewer you can put on any card
- Crews with a written mission and files of their own
- Loop cards, continuous or on a schedule
- Project memory shared by every agent
- Dictation and conversation mode
- Git window, editor, file manager and skills shelf
- Remote access from your phone
- Cost per pane, card and PRD, and a token dashboard
Checkout by Paddle, billed in your local currency with tax added where it applies. Cancel any time, and the licence runs to the end of the period you paid for.
Bring your own AI
The agents run on the subscriptions and keys you already have, and you pay the provider directly.
- A Claude subscription for Claude Code
- A ChatGPT plan with Codex
- GitHub Copilot for Copilot CLI
- An API key for opencode, Kimi, Qwen or any CLI that takes one
You only need the ones for the agents you use.
Runs on
Windows and Linux. There is no macOS build on sale yet.
What people ask before subscribing.
Not on the list?
Write to me and I answer myself.
contato@dcodertools.comDoes the subscription include model tokens?
No. You use the subscriptions and keys you already have and pay the provider directly, at its price. dcoder-terminal runs the work and shows you, in dollars, what each model spent.
Is this vibecoding?
Vibecoding is accepting whatever comes out. Here the work is specified before it starts, every card carries acceptance criteria, and a card can name a second agent that checks the result against them before it closes. Some people call a pile of parallel agents a swarm. The difference is that you wrote down what each of these has to do.
Does my code leave my machine?
Not to us. The board is a folder inside your project, plain Markdown and JSON, and the panes are processes on your machine. What goes out is what the agent you chose would send anyway, to the provider you configured. Features you turn on with your OpenAI key, such as commit messages and thread titles, send their input to OpenAI.
Do I need to be online?
To start, yes, because the app borrows a seat when it opens. After that it checks in every ten minutes and keeps working for up to 72 hours if it cannot reach us. If we answer that the subscription lapsed, the app says so and stops.
Can I use it on more than one machine?
Yes. Registering machines is free and has no limit. A seat is lent while the app is open and returned when you close it, so what counts is how many copies are open at the same moment.
Won't two agents in the same repository trip over each other?
They can, if you set it up that way. Dependencies hold a card in the Queue until the one before it closes, the scheduler runs only as many cards at once as you allow, and cards that run side by side are given separate files to touch. Branches and merges stay your decision.
How does the app know an agent finished or got stuck?
The agent tells it. Every terminal gets a small board tool, and the briefing at the start teaches the agent to mark its own state: working, waiting with a question, done. On first class CLIs the app also listens to the CLI's own end of turn signal.
Does it install or replace my agents?
Neither. It runs the CLIs already installed and signed in on your machine. If typing claude in a terminal works today, it works here. What it adds is its own board tool to each agent's config, in the narrowest scope it can.
Which services do dictation and conversation mode use?
Dictation transcribes with OpenAI or Groq, and conversation mode runs on OpenAI, both with your own key kept in the system credential store. The audio goes to the provider you picked, so the app tells you that in its settings.
What happens if I cancel?
The licence runs to the end of the period you paid for, then the app stops opening. Everything on disk stays yours: the board, the PRDs and the notes are Markdown files you can read with or without the app.
Install it once and it runs the agents you already have.
One app for every workspace. Your CLIs stay where they are.
Windows This computer
Download for Windows Linux This computer
Download for Linux There is no macOS build on sale yet.
Needs an active plan. Without a licence key the app opens on the activation screen and goes no further. See the plan