ForgeDeck

ForgeDeck One Mac window for every product you build

You run a SaaS, a client site, a plugin, an app. Each has its own repo, host, dashboard, and terminal tab. ForgeDeck puts them side by side: what's live in production and how that's known, what's broken right now, and one click into a terminal, your editor, or an AI agent in the right folder.

Free · native macOS · no account, no server, no telemetry · secrets stay in your Keychain

A ForgeDeck product page with demo data, with four labelled parts: products in groups, the evidence for each environment, the Play button, and where each fact came from.

What it's for

ForgeDeck is for the developer, indie maker, or small team that ships more than one thing. It answers the questions you'd otherwise answer by opening ten tabs.

The questionHow ForgeDeck answers itSheet
Is anything broken right now?One Needs attention list across every product: failing CI, endpoints that stopped answering, production falling behind, a disk filling up.1
Which version is live in production, and how do I know?Local, GitHub, Staging, and Production side by side, each with where the fact came from and how old it is. Never guessed from a green build.2
How do I run this project again?The Play button works out how it runs (dev server, simulator, Mac app, Docker, a local WordPress site) and starts it.3
Can an AI agent just fix this?Keep a to-do list per product and send items to Claude Code, Codex, Gemini, or OpenCode, started in the right folder with a map of the code.4
How do I start something new?Describe the idea. Your agent recommends a stack in a blueprint you approve, then builds it. The project arrives in ForgeDeck with Play, an agent, and its environments ready.5
Which CI run failed, and why?Every checkout's branch, uncommitted work, and CI status in one table. A failed run opens its log trimmed to the error, with a fix prompt.6
Did anyone sign up or pay today?Leads, signups, and payments from each product, read from sources you connect, each labelled with where it came from.7
Where's the staging API key?A Keychain vault scoped by product and environment. Commands get only the variables you allow; production asks for Touch ID.8

The tour

Every picture below is the real app, filled with eight demo products. Nothing is mocked up.

The ForgeDeck Overview with demo data: counts of active products, healthy environments, uncommitted codebases, failing pipelines, and new leads; a Needs attention list; product cards; and an activity feed.
Sheet 1 · Overview · demo data

See every product at once

Open ForgeDeck in the morning and know where things stand before you open anything else.

  • Needs attention What to look at first, across all products: production behind staging, CI failing on main, a status endpoint rejecting its token.
  • At a glance Healthy environments, codebases with uncommitted work, failing pipelines, and new leads this week.
  • Product cards Each product's local branch, GitHub state, and production version in one row, with buttons for its terminal, editor, and agent.
  • Groups SaaS, client work, internal tools: group them your way, star favorites, drag between groups.
  • Activity New subscriptions, releases going live, slow endpoints, failed builds, as they happen.
The Deployments page with demo data: a drift list of stages that differ, and a table of every product's local, GitHub, and production commits with the evidence for each: Verified, Reported, Stale, or Unknown.
Sheet 2 · Deployments · demo data

Know what's live, and how you know

"Deployed" should mean something. ForgeDeck never assumes production from a main branch, a green CI run, or a release tag. Each stage says how it's known.

  1. Localmain · cleanVerified · git on this Mac
  2. GitHubin syncVerified · GitHub API
  3. Stagingv2.15.0-rc.2Reported · not checked
  4. Productionv2.14.0Stale · 3 days old
Verified
ForgeDeck checked it: git, the GitHub API, the site's own version URL or status endpoint.
Reported
Something said so, like a deployment record. Not confirmed.
Stale
It was true. It's too old to trust now.
Unknown
No evidence, and the reason why.

Production is read from what you connect: a version URL, a status endpoint, GitHub deployments, an SSH host (read-only commands), Docker, or a release channel like WordPress.org, npm, or PyPI. Optional open-source companions for Node, Python, and WordPress report the exact running commit.

A product page in ForgeDeck with the built-in terminal open at the bottom, showing the header an AI agent gets: its context file, linked repositories, and permissions. Demo data.
Sheet 3 · Built-in terminal on a product page · demo data

Run it, open it, work on it

Every product is one click from running. No more digging through a README to remember the dev command.

  • WebStarts the dev server, opens the page when it answersnpm run dev → localhost:5173
  • MobileBuilds and runs on the iPhone simulator or an Android emulatornpx cap run ios
  • DesktopStarts the Mac, Tauri, or Electron appswift run · ./build.sh
  • SiteOpens a WordPress plugin's local site adminsite.local/wp-admin
  • Terminal In the product's folder, inside ForgeDeck or in Terminal, iTerm, Ghostty, or Warp. Drop a file or screenshot on it to type its path.
  • Editor Cursor, VS Code, Zed, or Xcode, in its own window for that project.
  • Agent Claude Code and others start in the folder with a compact map of the repo and any repos you link to it.
The Issues tab for a demo product: a to-do list of short notes, each with an Implement with Claude Code button, a GitHub issue number on one, and the Add to GitHub switch turned off.
Sheet 4 · Issues · demo data

Hand work to your AI agent

Jot a short note, like "CSV export drops the last row". Press Implement with Claude Code. The agent opens in that product's folder, reads the code, tells you what it thinks the note means, and asks its questions before it changes anything.

  • Issues per product A to-do list that lives on your Mac. Select several and the agent asks about all of them at once, then works through them one at a time with tests after each.
  • Any agent Claude Code, Codex, Gemini CLI, OpenCode, or Antigravity. Pick one in Settings.
  • The right context The agent gets a short map of the product and of related repos, not a wall of text. It reads more only when the work needs it.
  • GitHub, if you want Copy an item to a GitHub issue with your own gh login. Off until you turn it on.
  • AI usage A side panel shows tokens used and limits left, read from your agents' local logs. Nothing is uploaded.
The blueprint review for a new demo project: the agent recommends Next.js with PostgreSQL and Stripe and says why, with two alternatives, the planned structure, and which tools are installed. Nothing is built until it is approved.
Sheet 5 · Reviewing a new project's blueprint · demo data

Start new, or let the agent set up

New Project turns an idea into a working project. ForgeDeck makes the folder, the Git repository, and the product; your agent recommends a stack in a blueprint. You approve it, and the agent builds it. For an existing repository, ask once and the agent works out the setup. Either way it writes to a file, ForgeDeck checks it, and you approve before anything changes.

  • New projectStack, structure, build steps, production plan, variable names.forgedeck/blueprint.json
  • EnvironmentsWhere it's deployed and how to read the live version.forgedeck/environments.json
  • RunHow it runs, tried first, filled into the Play button.forgedeck/run.json
  • LeadsWhat it collects, and which of those to trackforgedeck leads
  • SkillsAgent skills and plugins that fit the product.forgedeck/skills.json

The agent only proposes and builds locally. It never deploys, pushes, or sees a token, and installs no skills itself. A planned production host stays Planned until a check answers.

The Git and CI page: every checkout with its branch, working tree state, commits ahead and behind, and CI status, one of them failing. Demo data.
Sheet 6 · Git & CI · demo data

Catch what's uncommitted or failing

Every checkout and worktree on your Mac in one table: branch, uncommitted work, ahead or behind, and whether CI passes.

  • Read from disk Local state comes straight from git and works offline.
  • CI failures, explained A failed GitHub Actions run opens its log trimmed to the error, and writes a fix prompt to hand to your agent.
  • Releases Which version is published on WordPress.org, npm, PyPI, or GitHub, next to the version in your code.
The Leads and Metrics page: leads, signups, paid, and cancellations for the last 7 days with trends, a funnel, and a table of every number with its source and age. Demo data.
Sheet 7 · Leads & Metrics · demo data

See what your products bring in

Leads, signups, payments, and cancellations from every product in one place, so a quiet contact form doesn't go unnoticed for a week.

  • Only real sources Numbers come from endpoints you connect. Each shows its source and age; a source that stops answering turns Stale rather than dropping to zero.
  • Set up by your agent It finds what the product collects (forms, signups, orders, bookings), asks which to track, and builds the check.
The Secrets Vault: variables scoped globally, by product, and by environment, with Keychain values hidden and production marked; a preview of what a process would receive in Staging. Demo data.
Sheet 8 · Secrets Vault · demo data

Keep secrets out of .env files

API keys and database URLs live in the macOS Keychain, scoped to a product and an environment, and go only to the commands you allow.

  • Scoped Global, product, environment, or profile. See exactly what a process would receive before it runs.
  • Production asks first Revealing or using a production secret asks for Touch ID.
  • Safe for agents forgedeck exec runs a command with the right variables. Agents see names, never values.
The command palette open over the Overview, listing products to open and actions to run.
Sheet 9 · Command palette · demo data

Keyboard first, and a command line

Press ⌘K to open any product, start its agent, run an action, or jump to a page. Every button has a palette command.

  • The forgedeck command The app is also a CLI: list, status, exec, context, ci, usage. Your agents use it too.
  • Remote Forge Remote keeps Claude Code's Remote Control running, so you can work on any product from the Claude app on your phone.

Privacy and safety

  1. Everything stays on this Mac. No account, no analytics, no server.
  2. Secret values live only in the macOS Keychain. They are never written to ForgeDeck's files, logs, exports, or command arguments.
  3. Variables are scoped to a product and environment and go only to the commands you allow. Production secrets ask for Touch ID.
  4. External systems are read-only by default. Anything that writes remotely or touches production names its exact target and needs a press and hold.
  5. SSH runs a fixed list of read-only commands, with strict host key checking.
  6. The only automatic network request is a daily check for a new release on GitHub. It sends nothing about you, and it can be turned off.

Install

  1. Download the zip from the latest release and unzip it.
  2. Drag ForgeDeck.app into Applications.
  3. Open it. ForgeDeck is signed with a Developer ID and notarized by Apple, so it opens like any other app.
  4. Add a product. A folder, a GitHub repository, an SSH host, or just a name. Or start with the demo products to look around.

Updates install themselves, after ForgeDeck checks that the new copy is signed by the same developer.

Revisions

RevDescription
0.13.7Select any number of issues, GitHub ones included, and plan or implement them in one agent session. Finished plans move their issues to Done, and done GitHub issues can be closed from ForgeDeck. Quitting closes ForgeDeck all the way, ending every terminal's processes.
0.13.6Issues include the product's GitHub issues: open ones show under On GitHub, and Add to List or Implement brings them to your agent. Add issues from the command line with forgedeck issue add, and list them with forgedeck issue list.
0.13.5Each product can start its own AI agent and model. Restart Agent loads newly installed skills into a running session. AI Usage shows Antigravity (agy) limits and flags paid weekly usage that resets unused within a day.
0.13.4Products have a Memory: rules and key facts every agent session reads first. Agents add what they learn, and you edit or delete anything in the Memory tab.
0.13.3Add a GitHub repository and clone it in one step: Add Product shows the folder it goes into, and the clone becomes the product's local folder. An existing folder is never overwritten.
0.13.2AI Usage shows Claude Code's plan limits: the 5-hour window, the weekly window, and weekly limits per model, with reset times and your plan, checked through Claude Code the way /usage does.
0.13.1Forge Remote: every session you start from the Claude app starts clean, with no long-lived session that piles up old context. It starts reliably when ForgeDeck reopens, stopping a server left from the last run instead of failing.
0.13.0Plan a feature with your agent: review its steps as a checklist, then execute, revise, or discard. Pick a product by number or part of its name with forgedeck find, or /products and /product in Claude Code and Forge Remote, which now load a product without asking for approval. Issues the agent finishes move to Done with its summary. New projects get skill suggestions with the blueprint.
0.12.0New Project: describe an idea and your agent plans the stack in a blueprint you review and approve, then builds it. Or pick a starter (Next.js, FastAPI, Swift, Go, Rust, WordPress, and more). A setup checklist tracks every step and picks up where it stopped. Set up leads for any product from the Leads page.
0.11.2Select several issues and implement, finish, or delete them together; the agent asks about all of them first. Skill suggestions appear the moment the agent writes them. Drop files, folders, or screenshots on a built-in terminal to type their paths.
0.11.1Issues per product: a to-do list where Implement with your agent opens the item in a terminal, and the agent asks what it needs to know before it starts. Signed with a Developer ID and notarized. Claude Code no longer asks to trust the context folder every session.
0.11Skills per product: your agent suggests skills and Claude Code plugins that fit the work. Install them with one click; active ones are named in the agent's prompt.
0.10Set Up Leads: your agent finds what a product collects, asks which to track, and sets up the check. forgedeck leads confirms leads are arriving.
0.9Set Up Run: your agent works out how a product runs, tries it, and proposes the commands for one review.
0.8Repositories propose their environments, production included, for review. Nothing becomes production without your approval.
0.7Forge Remote: work on any product from the Claude app. Agent context loads only what a task needs.
0.6Version URLs and a Releases card for WordPress.org, npm, PyPI, and GitHub.
0.5Agent skills in the Toolbox, repository tools, and MCP servers per product.
0.1–0.4Products, the evidence chain, Keychain secrets, the forgedeck command, the built-in terminal, CI failure logs, AI usage, Play, and automatic updates.