Build with Helmdeck.
Guides and reference for orchestrating your AI coding tools into automated pipelines.
App Overview
Helmdeck is a native Mac app that unites the AI coding assistants you already use — Claude Code, Gemini, Codex and more — into one organized home where you plan, run and watch automated work, without juggling a stack of separate terminals.
Working with AI coding tools today means bouncing between separate terminal windows, scattered config files and disconnected apps — losing track of which agent is doing what, and watching your context evaporate every time you switch tools. There is no single place to see your skills, sessions, automations and project boards together. Helmdeck collapses that sprawl into one window, so your work and its history finally live in one place.
- Brings several AI coding assistants together under one roof, showing only the tools you actually use.
- Organizes everything through a slim activity rail and panel — Boards, Skills, Pipelines, Contexts, Projects and Sessions, all a click apart.
- Runs and tracks multi-step automations, including designing a whole pipeline from a single plain-language request.
- Gives each project its own board with a command-center overview, plus Kanban, Canvas, Cloud, Workspace and Settings views.
- Embeds live terminals inside the app, with floating bubble cards so background runs stay glanceable without stealing focus.
- Adapts to you with light, warm and dark themes and full English and Spanish localization.
- On first launch, a quick guided setup and a brief skill scan get you ready.
- Pick a section from the activity rail; the panel beside it shows just that content.
- Open a project board and switch between its modes — overview, Kanban, Canvas and more.
- Kick off AI work; pipelines and terminals run inside the app and collapse into bubbles you can watch while you keep working.
- Adjust tools, models, theme and language anytime from one settings panel.
- Local-first: your skills, pipelines, sessions and contexts live on your own Mac, and pipeline outputs are written to a local folder you own.
- Works alongside the AI tools you have already installed, watching their files and reflecting changes automatically.
- Each provider ships with one sensible default model, and you can add your own per tool — new models are validated before they are used.
- Tools you leave disabled are fully ignored — they never touch disk or clutter your view.
Skills
Every saved instruction and prompt your AI tools can follow — Claude Code, Gemini, Codex and more — gathered into one searchable library, so a skill you write once works everywhere instead of staying trapped inside a single tool.
Each AI assistant hides its skills away in its own folders and its own format. The same useful instructions end up scattered across tools, silently duplicated, and impossible to reuse without copying files by hand and keeping them in sync. Helmdeck pulls them all into one place and keeps shared skills linked — so improving a skill once improves it for every tool that uses it.
- Automatically discovers every skill already installed across your AI tools — both the global ones and those scoped to a single project.
- Recognizes a skill whether it is a single file or a folder, and reads its name and description for you.
- When the same skill is shared by several tools, shows it once — tagged with every tool that uses it — instead of cluttering your library with duplicates.
- Create a new skill from ready-made starter templates tailored to each tool, or edit any skill in a built-in editor that saves straight back to disk.
- Group skills into collections, mark favorites, and full-text search across names, descriptions and content.
- Discover community skills from an online library and install one to several tools at once.
- Helmdeck scans your tools’ folders and lists every skill it finds in one sidebar.
- Filter by tool, favorites or collection — or search across everything at once.
- Open any skill to read or edit it, or create a new one from a template; your changes save right where the tool expects them.
- To add community skills, search the online library, pick which tools to install them for, then refresh to re-scan.
- Scanning runs when you open the app or hit refresh — there is no live file-watching, so refresh after editing skills outside Helmdeck.
- A skill’s name is how your tool invokes it, so Helmdeck tidies names (lowercase, dashes) when you create one.
- Global skills stay separate from project-specific ones, and agents are kept visually distinct from ordinary skills.
Sessions
A single, searchable home for every past AI coding conversation — across Claude Code, Gemini, Codex, Qwen and your own custom sources — neatly organized by project.
Each AI assistant stores your conversation history in its own hidden folder and its own file format, so revisiting a useful past session means digging through scattered directories you cannot easily browse. Finding the right conversation, remembering what was discussed, or picking up where you left off is slow and error-prone. Helmdeck reads all of those native histories in place and presents them together, so nothing stays trapped in any one tool.
- Gathers conversations from every enabled tool automatically and groups them by the project you were working in.
- Merges a project that appears across several tools into one entry with a combined session count, so the same folder is never listed twice.
- Rename any session with a friendly title that sticks, and search both across sessions and within a single conversation, with matches highlighted.
- Renders the full transcript richly — tool actions, file edits as color-coded diffs, command output and pasted images.
- Favorite the ones you reach for often, and multi-select delete with a 10-second undo (deleted sessions go to the Trash, never lost).
- Updates live as conversations change on disk, even when you rename or continue them outside Helmdeck.
- Browse the project tree in the sidebar; each project shows how many sessions it holds.
- Open a project to see its sessions, newest first, loaded instantly on repeat visits.
- Click a session to read the full transcript; search, scroll, or jump between matches.
- Rename it, mark it a favorite, or hit “Open in Terminal” to resume the conversation exactly where it stopped.
- Only tools you have enabled are scanned; disabled tools’ projects stay hidden.
- Custom-source sessions can be browsed and read, but cannot be resumed in the terminal — there is no backing tool to continue them.
- Renames are non-destructive and reflected back, so the underlying tool shows the new title too.
- Long histories load in pages, keeping even huge projects responsive.
Live Terminal
Run real AI coding assistants right inside Helmdeck — Claude Code, Gemini, Codex, Qwen — with a live status badge that shows, at a glance, exactly what each agent is doing.
Run an AI coding tool in a normal terminal and you are staring at a wall of scrolling text, with no clear sense of whether it is thinking, reading your files, making changes, stuck on a question, or finished. Juggling several agents means juggling several terminal windows, each demanding your attention. Helmdeck brings these tools into one place and translates their raw activity into a clear, glanceable signal.
- Runs the actual command-line tools natively, exactly as they would run in a real terminal — full color, fully interactive.
- Shows a live status badge for each session: idle, thinking, reading files, working, waiting for you, error, or done.
- Pairs every session with an animated agent character whose state mirrors what the tool is doing right now.
- Detects when an agent pauses to ask a question and surfaces the choices as a clean, clickable menu instead of raw prompts.
- Runs multiple sessions side by side in tabs, each with its own working folder and status.
- Finds your installed AI tools automatically, with no manual PATH setup.
- Open a tab and pick the AI tool you want; Helmdeck launches it in a real terminal session.
- The status badge and animated agent update continuously as the tool thinks, reads, edits or runs commands.
- If the tool asks for a decision — trust this folder, approve a change, pick an auth method — it appears as a numbered, clickable menu.
- Pick an option and Helmdeck answers the tool for you; when the task finishes, the badge flips to “done.”
- The status badge is a read-only signal — it never alters or interferes with what the agent does.
- Status detection is tuned per tool, so the badge reflects each tool’s real prompts rather than guessing.
- Drive sessions hands-on — typing, arrow keys, confirmations — or let Helmdeck run them more automatically.
Contexts
Reusable project notes — your standards, key facts and goals — that get woven straight into your automations, so your AI tools always start with the right background.
Every time you ask an AI coding tool to do something, you end up re-explaining the same things: how the project is structured, what conventions to follow, what you are ultimately trying to achieve. That repetition is tedious, and skipping it produces inconsistent, off-target results. Contexts capture that knowledge once, so every automation starts from the same shared understanding.
- Save reusable notes — coding standards, architecture facts, goals, glossaries — as plain documents tied to a specific project.
- Injects the right notes directly into your automation prompts, so the AI always has the background it needs.
- Keeps notes scoped per project, and organizes them into folders so larger projects stay tidy and searchable.
- Optionally surfaces the context files your AI tools already keep in the project, alongside your own notes, as read-only references.
- Import existing notes individually or by whole folder, preserving their structure.
- A full viewer and editor with preview, search, copy and reveal-in-Finder.
- Open a project’s Contexts area and add a new note or import existing files.
- Write or paste the background you want the AI to always know.
- Reference that note inside an automation step, so its content is woven into the prompt.
- Run the automation — the AI now works with your standards and goals built in; edit a note anytime and updates flow into future runs.
- Notes are scoped per project, so they never leak across unrelated projects.
- Context files discovered from your tools’ own folders are viewable but managed separately — only your saved notes get injected into prompts.
- Duplicate detection hides a note that already exists, so the same file never shows up twice.
- Deletions are reversible within a brief undo window before anything is permanently trashed.
Pipelines (Automations)
Saved, multi-step automations that run your AI assistants one after another, capturing every step’s result as a document you can read, reuse and feed into later steps.
Real work with AI tools rarely fits in a single prompt — you analyze, then generate, then test, then document, often switching between assistants. Doing this by hand means babysitting one terminal, copy-pasting outputs between steps, and losing track of what each tool produced. Pipelines turn that whole chain into one repeatable, hands-off run with every result preserved.
- Runs a sequence of steps, each one an AI tool working in your chosen project folder.
- Lets each step use a different tool and model, so you can match the right assistant to the right job.
- Saves every step’s output as its own document on disk, alongside the exact prompt that produced it.
- Lets a later step reference earlier results — the previous output, a specific step, or all prior outputs — so work flows forward without copy-paste.
- Chains steps to run automatically, or pauses for your review at the sensitive points.
- A per-step permission toggle auto-approves safe, read-only work while keeping a human in the loop for risky actions.
- Create a pipeline pointed at a project folder and add steps.
- For each step, pick the tool, model, prompt and any skills, and decide whether it auto-runs or waits.
- Launch it; an embedded terminal opens and you watch the first step run live.
- Each step writes its result, signals completion, and the next one starts on its own — read any saved output, edit it, or open the project in Finder.
- Each pipeline and all its outputs are stored as plain files you own.
- Minimize a running pipeline into a small bubble and keep working while it runs in the background.
- The last step never auto-advances, so the final result always lands in front of you.
- Stopping a running step asks for confirmation first, so long jobs are not killed by accident.
Smart Pipeline Generator
Describe what you want in plain language and an AI planner designs the entire multi-step pipeline for you — choosing the tool, model, prompt and skills for each step, and explaining why.
The hardest part of automation is staring at a blank pipeline and not knowing how to break a big goal into the right sequence of steps. Deciding which assistant handles which part, in what order, and how the outputs connect takes real expertise. The Smart Pipeline Generator does that decomposition for you from a single description of your objective.
- Turns a natural-language objective into a complete, ready-to-run pipeline.
- Picks the tool, model, prompt and skills for every step, and orders them sensibly.
- Decides which steps can run automatically and which should pause for your review.
- Lets you attach images, reference files (typed inline with @) and saved project context to the objective.
- Only offers the AI tools you have actually enabled, so it never plans around something you do not have.
- Wires step outputs together automatically, and records a plain-language rationale for the plan and each step.
- Click the sparkles button, choose your project folder, and pick which assistant should do the planning.
- Describe your objective, pasting images or referencing files and saved context as needed.
- Click Generate; the planner runs in a live terminal you can watch — or minimize it to a bubble and keep working.
- When it finishes, the proposed pipeline is assembled with its reasoning attached, ready to open, review and run.
- A history of past attempts lets you retry a failed run or continue one with extra instructions.
- If a run produces only a partial result, a Recover option can salvage it without starting over.
- Clear error messages explain failures — for example, a tool that needs sign-in or a model that was not found.
Board (Kanban)
A visual board where every card is more than a note — each one carries a real, end-to-end AI automation you can see, steer and run, in Kanban, Timeline or List view.
AI coding assistants are powerful but unstructured — you fire off one prompt at a time and lose track of what is planned, running or done. The Board turns loose AI work into an organized pipeline you can see and steer, so a backlog of ideas becomes a visible flow of automated work with clear status at every stage.
- Gives each project its own board with cards that each carry a real, runnable AI automation — not just text.
- Offers three card types: a standard objective card, a Fix card that diagnoses a bug before touching code, and a Smart card that splits a big goal into linked sub-tasks.
- Generates a card’s concrete step-by-step plan by handing your description to the assistant of your choice.
- Tracks live status on every card — generating, ready, running, paused, done or failed — with inline progress and per-step output you can open.
- Lets you refine work mid-flight: re-review to add steps, or re-generate to adjust and replace the plan without losing history.
- Supports Kanban, Timeline and List views over the same cards, plus custom columns, epics, cycles, tags, priorities and search.
- Create a card in Backlog, describe the goal in plain language, and attach images or reference files if useful.
- Drag it to Generate — the assistant drafts the concrete steps; you confirm before it runs.
- Drag to In Progress to execute the steps; watch progress and open any step’s output.
- Move to Review to inspect results, then Done when satisfied — or drag back to re-review and add more.
- Pipeline generation is driven by dragging a card into the Generate column — the board’s columns are the controls.
- A Fix card pauses for your approval of a written diagnosis before any fix is generated.
- Card status is always computed live, never stale, and a detachable pop-out lets you edit a card or read output without blocking the board.
- Each board carries its own accent color to keep your projects visually distinct.
Isolated Git Workspaces
An optional safety layer that runs a card’s automation inside a separate, isolated copy of your project — on its own branch — instead of your live working files.
Letting an AI make sweeping edits directly on your main code is risky — a bad run can leave your checkout in a messy, half-changed state. Isolated workspaces give each card its own private copy of the repository on its own branch, so experiments and automated edits never touch what you are actively working on until you choose to bring them in.
- Creates a private, isolated copy of your project for a card, on its own automatically named branch, when you opt in.
- Keeps these copies in a dedicated managed location, so your real checkout is never modified.
- Runs every step of the card’s automation inside that isolated copy.
- Shows you, before the run starts, whether the work will be isolated and exactly which branch it will use.
- Extends the same isolation to Smart sub-tasks, deriving each one from the parent card’s branch.
- Turn on the workspace option for a card (or globally) before running it.
- When you start the card, a confirmation shows whether it will be isolated and the branch name it will use.
- Approve, and the isolated copy is created and the automation runs there.
- Review the results, then delete the workspace when finished — with a 5-second undo if you hit it by mistake.
- It is opt-in per card — by default work runs in your main project, and you choose isolation deliberately.
- If a workspace folder ever goes missing, the app detects it and quietly falls back to the main project.
- Cleanup removes the folder first but keeps the branch briefly, so undo restores everything intact — and a crash during cleanup never loses your work.
MCP Catalog
A single, read-only directory of every external connection (MCP server) your AI tools can reach — assembled automatically from the setup you already have.
When you run several AI assistants, each keeps its own list of external connections — databases, web services, file stores and more — buried in separate configuration files. There is no single place to see what your tools can actually touch. Helmdeck gathers all of it into one view, so you always know what is connected and where it came from.
- Lists every MCP server your installed AI tools are configured to use, all in one place.
- Groups connections by the tool that owns them, or lets you focus on a single tool at a time.
- Shows each connection’s status — configured, currently connected, or needing sign-in.
- Labels how each connection is reached (local command, web endpoint, remote service or built-in) and where it came from.
- Distinguishes user-wide connections from project-specific ones.
- Search across names, tools, status, endpoints and the specific capabilities a remote connection offers.
- Open the MCP section to see “All MCP” plus a per-tool breakdown.
- Helmdeck reads your existing tool configurations on the Mac and builds the list live.
- Browse by tool group, or pick one tool to narrow the list.
- Select any connection to see its details — what it links to, how, and its current state — or search to jump straight to one.
- Read-only: Helmdeck surfaces what is already configured; it does not create or change connections.
- Only tools that are both enabled and installed on your machine appear.
- The list refreshes itself, so it always reflects your current setup.
Notifications
A built-in activity inbox that records every meaningful event in Helmdeck, paired with optional Mac banners so you stay informed even when you have stepped away.
Pipelines, context syntheses and smart-plan generations can run for a long time, and they often finish while you are focused elsewhere. Without a reliable record, it is easy to miss when work completes, a step fails, or a task needs your input. Helmdeck keeps a persistent log and can ping you the moment something important happens.
- Captures key events — steps completing, pipelines finishing, failures, smart-plan results, saved contexts, and steps that need your input.
- Keeps an always-on in-app inbox you open from a bell in the sidebar.
- Shows a red dot and unread markers so you can spot new activity at a glance.
- Optionally raises native Mac banners — with sound — for the events you care about.
- Lets you choose which event types may notify you, with a master on/off switch.
- Color-codes and icons each event by type for quick scanning; mark items read, mark all read, or clear the inbox.
- As you work, Helmdeck logs each notable event to your in-app inbox automatically.
- The first time, it asks macOS permission to show system notifications.
- When an enabled event fires, you get a Mac banner (if you allowed it) and an inbox entry.
- Click the bell to open the inbox, review items, and mark them read — or adjust which types are allowed to notify you.
- The in-app inbox always records events, even if you decline Mac banners or switch categories off.
- Banners appear even while Helmdeck is in the foreground.
- The inbox keeps your recent activity and persists across restarts, fully in English and Spanish.
Workspaces & Sharing
Shared team workspaces where boards, rules and automations stay in sync live across every member’s Mac — so a good improvement spreads to the whole team instantly.
When a team uses AI coding assistants, everyone ends up maintaining their own private rules and project guidance, so good practices never spread and quietly drift out of sync. Workspaces give a team one shared layer on top of each person’s app — when one member improves something, everyone benefits, with no copy-pasting conventions between teammates.
- Create a shared workspace and add teammates with roles — owner, admin or member.
- Share project rules item by item: you choose exactly which rules to publish, and others subscribe one at a time.
- Propagates an edit instantly to only the people who subscribed to that specific rule.
- Lets owners and admins manage members and remove shared items for the whole team.
- Binds one workspace to one board, so a board becomes a clear, scoped team space.
- Keeps a copy on each Mac, so your shared content stays readable even offline.
- Sign in, create a workspace, and invite teammates.
- Click Share on a rule you want the team to have; subscribers pick it up in their workspace view.
- Edit a shared rule and the new text fans out live to its subscribers.
- If you ever leave or sign out, your shared rules quietly convert back to private local rules, text intact.
- Permissions are enforced on the server, not just hidden in the app, so access is real rather than cosmetic.
- Authors can edit their own rules; only admins and owners can delete shared ones.
- Sharing requires being signed in; everything else works locally without an account.
Cloud Folders
Secure shared file folders inside a workspace, where the folder’s owner controls exactly who can see and download each one.
Teams need to pass real files — specs, exports, assets — alongside their shared rules and boards, but most sharing is all-or-nothing or quietly leaks files to people who should not see them. Cloud Folders give each folder its own private guest list, so the right people get the right files and no one else, with access checked on the server every time.
- Lets any workspace member create a folder and upload or download files into it.
- Gives the owner per-folder control — share with specific people, with whole roles, or with everyone.
- Shows clear privacy indicators (private vs. shared) plus who a folder is shared with.
- Tracks storage use against a generous quota, with sync status and file counts.
- Runs transfers in the background, so big uploads keep going while you work elsewhere in the app.
- Protects against mistakes with confirm-and-undo on deletes.
- Open the Cloud area on a board linked to your workspace.
- Create a folder and drag in files, or open one shared with you to download.
- As the owner, open the share editor and pick the roles or members who may see it.
- Watch progress in a floating transfer chip that follows you across the app.
- Folders are upload/download only — there is no in-app preview.
- A hidden member cannot see a folder at all; visibility is enforced server-side, so the app only ever shows what you are truly allowed.
- File deletes briefly delay so you can undo, since cloud files cannot be restored once gone.
Setup & Sign-in
A friendly first-launch setup that gets you running locally in minutes, plus an optional sign-in that unlocks team collaboration when you want it.
Powerful developer tools often demand accounts, permissions and configuration before you can do anything. Helmdeck flips that: it walks you through a short guided setup, lets you start with no account at all, and only asks you to sign in when you want the features that actually need it. AI tools start off until you deliberately turn them on, so nothing runs behind your back.
- Guides first launch through a short series of steps — language, appearance, default AI tool, notifications and more.
- Starts every AI assistant turned off and lets you enable each one explicitly, checking install, sandbox and login status as you go.
- Offers sign-in with Apple, Google or email — or “continue without an account.”
- Stores your session securely in the system Keychain and restores it automatically next launch.
- Supports password recovery through a secure email link, and lets you set a profile photo shown across the app.
- On first open, step through the setup sheet and pick your preferences — changes apply live.
- Turn on the AI tools you use; Helmdeck verifies each is installed and signed in.
- Optionally sign in to unlock workspaces and cloud folders, or skip it and keep working locally.
- Your session is remembered, so you land straight in the app next time.
- Guest mode works with no account; sign-in is only needed for collaboration.
- Sign-in uses native Apple and Google flows, with no third-party tracking.
- Sign-up never reveals whether an email already exists, protecting your privacy.
Languages & Updates
Full English and Spanish support with instant switching, paired with secure, signed updates so you are always on the latest release.
Bilingual teams should not have to restart an app or hunt through menus to change language, and no one should have to wonder whether an update is genuine. Helmdeck switches languages instantly and keeps every screen translated, while updates arrive securely from a trusted, verified source.
- Offers the entire interface in English and Spanish.
- Switches language instantly, with no app restart, applying everywhere at once.
- Lets you choose your language right in the first-launch setup, and change it anytime.
- Keeps translations organized by feature, so every screen — including newer ones — stays fully covered.
- Delivers app updates that are securely signed and verified before installing.
- Pick your language during onboarding, or change it later in settings.
- The whole app re-renders immediately in your chosen language.
- When a new version is available, it is fetched from the official update source.
- The update is verified as authentic before it is applied.
- Your language preference is remembered between launches.
- Translations are split into focused groups, so new features arrive fully localized rather than half-translated.