The Construct

A complete walk-through: every room, every feature, click by click.

User guide · version 0.1.50 · for founding members

What The Construct is

The Construct is your team, in one app. Instead of one chatbot, you have a small roster of named people, each with a role, a voice, and a job. You talk to them, they think with you, and they go do work and bring it back. You sit in the CEO chair and decide what matters; the team carries the rest. Now you have people.

There is one person you will talk to most: Morpheus, your thought partner. Around him is the roster (Trinity, Tank, Link, Mouse, the Operator, the Concierge, and any others you have set up), plus the Architect, who turns a request into a coordinated plan across several of them.

Always on the frontier. Part of having a team is not having to track the AI field yourself. The Construct routes your agents to the most capable models you have connected, starts new work on the current frontier default (today, Claude Opus 4.8), and when a stronger default ships it moves your agents to it and tells you it did. You set the direction; keeping up with what is newest and best is its job, not yours.

The one idea that explains the whole app

Everything is organized around where work belongs:

Voice runs through all of it. You can speak to the app from anywhere, and it answers in each person's own voice. The rule the app follows: voice carries the trigger and a short answer; the substance lands on a screen you can read. It will never read a long report aloud unless you ask it to.

Prefer doing to reading? This guide is the room-by-room reference. Its companion, Start the Adventure (the Playbook), is the trail: seven small quests with copy-paste briefs that take you from first hello to running a real project. Most people enjoy walking the trail first and keeping this guide within reach.
Two speeds, one thing to keep straight: when you are just talking with Morpheus, it opens, browses, fetches, and shows you things freely, and is instructed to stop short of spending, credentials, or anything it cannot undo. When you put a plan in motion, your agents are built to execute it, including steps that cost money, under the gates you set (plan approval, involvement level, cost ceilings). Knowing which mode you are in is the whole trick. More on this at the end.

Getting around: the four rooms

The Construct has four rooms. You move between them with the small row of room icons in the bottom-left corner (and from the sidebar when you are in the Terminal).

Lobby · your front door. State what you need and the app routes it to the right place.
Dojo · a calm space to think out loud with Morpheus.
Decision Room · the short list of things waiting on your call.
Terminal · the workshop, where agents run and you watch the work.

To switch rooms

  1. Find the row of small icons in the bottom-left corner of the window. There is one for each room.
  2. Click the room you want. The screen changes immediately; nothing is lost in the room you left.

Good to know: You do not have to navigate by hand. From the front door or by voice, just say what you want and the app sends it to the right room for you.

The Lobby: your front door

The Lobby is where you arrive: a calm landing screen ("the Construct, come sit in a different chair") with the entrances to the app along the bottom. It is the quietest room. From here you step into the place you want to be.

The Lobby landing screen with the armchair and room doors
The Lobby. The doors along the bottom take you into each room.

To enter a room from the Lobby

  1. Look along the bottom of the Lobby for the labeled doors: Dojo, Terminal, and Decisions.
  2. Click the one you want. You land in that room.

You do not have to come through the Lobby to act. The fastest way to start something is the "What do you need?" prompt in the Terminal (covered below), or your voice: press Alt+Space from anywhere and just say it.

The phone on the side table in the Lobby art is a quiet nod to a real feature: The Construct can answer your actual phone by text and voice. See "Your phone," near the end.

The Dojo: thinking with Morpheus

The Dojo is a calm, single-column space for thinking out loud with Morpheus. It is not a task board. It is where you reason through something, capture an idea, or ask Morpheus to show you something or pull a quick fact. Morpheus greets you when you arrive and answers in his own voice.

The Dojo with Morpheus
The Dojo. Morpheus greets you; type or speak in the bar at the bottom.

To talk with Morpheus

  1. Click the Dojo room icon (bottom-left), or open it from the Lobby.
  2. Type in the "Type a message" bar at the bottom and press Enter. Morpheus replies in the single column above.
  3. To speak instead, click the microphone icon in the input bar and talk. Your words fill the bar; it sends when you finish.

Show me, and tell me: Morpheus can act, not just answer

Morpheus is not limited to talking. He can put a page on your screen and pull live facts.

Good to know: Morpheus only does this for quick shows and live facts. A real build ("make me a brand kit") still goes to the Architect, and a system task ("move my files") still goes to the Terminal. You do not have to think about which is which; just ask.

Attach a file to the conversation

  1. Click the paperclip icon in the input bar.
  2. Choose a file (or a folder). Its contents come into the conversation so Morpheus can work from them.

Convene the council from the input bar

The ⚖ button on the input bar turns on council mode: your next message convenes the LLM Council instead of going to Morpheus alone. The panel deliberates inline, and the finished verdict stays in the thread.

Hear a reply, and control the voice

  1. Each of Morpheus's replies can be played aloud. Click the small speaker next to a reply to hear it in his voice.
  2. To mute or manage voice, click the speaker icon in the top-right to open the voice menu.
Voice control menu
The voice menu (top-right): mute, and voice controls.

Smooth edges. When Morpheus speaks, the audio opens with a beat of silence before the first word and holds a beat after the last, so the voice never clips on or off at the edges.

Find an earlier conversation

  1. Click the conversations icon in the top-left to open the history panel.
  2. Click any past conversation to reopen it, or start a fresh one. Entering the Dojo always opens a clean space; your prior threads are one click away.
Dojo conversation history
The history panel slides in from the side. Reopen a past thread or start new.

The "Return to last conversation" pill appears briefly when you enter the Dojo, as a one-click way back to where you were.

Voice: hands-free from anywhere

You do not have to be in the Dojo, or even looking at the app, to use your voice. One key opens the voice door from any room.

The voice door overlay
The voice door, opened with Alt+Space. Speak, and it routes to the right person.

To talk to the app from anywhere

  1. Press Alt+Space. A small listening panel appears, wherever you are.
  2. Speak your request: "Trinity, pull the launch copy into the Decision Room", "what's the weather", "show me last night's scores", or "what needs my call?".
  3. The app understands who or what you meant and acts. It answers with a short spoken line, and the substance lands on the right screen. It does not read long things aloud.

What you can do by voice

Talk to one specific person

  1. Open Settings → Personalities.
  2. Click the "Talk" button next to the person you want. The voice door opens already addressed to them, and they answer in their own voice.

Each person sounds different. Every agent has its own voice, set in Settings → Personalities. When a reply is spoken, it is in that person's voice.

What voice does on its own: when Morpheus is answering you directly, it opens, browses, searches, and shows you things, but is instructed to stop short of buying, logging in, or submitting on your behalf, and to lay out the options instead. Work you hand off by voice ("Tank, build...") goes to your agents, which execute it under the gates you have set. See the safety line at the end for the full picture.

The Terminal: the workshop

The Terminal is the main work surface of The Construct. It is where your agent threads live, where you read and send messages, and where the Cockpit shows you the state of every active project and agent. When you want to do work, this is the room.

What the Terminal contains

The Terminal has two panels side by side: the Sidebar on the left and the main area on the right. The Sidebar lists all your threads and gives you quick access to cost, files, and skills. The main area shows either the Cockpit (a status dashboard) or the conversation you have open.

Getting there

  1. Click "Terminal" in the room navigation (bottom-left of any screen, or the door in the Lobby).
  2. The Terminal loads with the Cockpit open by default when no thread is selected.
  3. You can also click "Open Terminal" in the Sidebar header at any time. This closes any open thread and puts the Cockpit front and center.

Good to know: Press Cmd+B to collapse or expand the Sidebar. When collapsed it becomes a narrow icon rail; click the expand arrow at the top to restore it.

The Cockpit

The Cockpit is the status dashboard inside the Terminal. It has three zoom levels: a personal briefing (home), a project grid, and a system view of every agent and provider. Switch between them with the tab strip in the top-right of the Cockpit header.

The Cockpit with its three zoom tabs and a grid of project cards
The Cockpit. Top-right: the Briefing, Project View, and System View tabs. The main area shows your projects as cards.

Switching zoom levels

  1. Open the Cockpit by clicking "Open Terminal" in the Sidebar header, or by closing any open thread.
  2. In the Cockpit header, find the tab strip on the right: Briefing, Project View, and System View.
  3. Click any tab to switch to that view. The active tab is highlighted, and a "Updated just now" timestamp sits to its right.

A note on labels: these tab names change with your Personality Level (Settings, General). At the highest level they read "Simulations" and "The Source" instead of "Project View" and "System View." The views are the same.

Briefing (the home view)

The Briefing is your personal summary: what needs your decision right now, what changed since you were last here, and quick links back into recent threads.

  1. Click the "Briefing" tab.
  2. Read the State of Play at the top: how many decisions are pending and when activity last happened.
  3. Under What Needs You, each pending decision is a button. Click one to jump to the Decision Room and act on it.
  4. Under Since You Were Here, see the most recently active threads with a short preview.
  5. Under Jump Back In, click a thread pill to return to that conversation.
The project grid showing project cards with status and progress
Project View. Every active project as a card, with status, thread count, and progress.

Project View

Project View shows every active project as a card. Use it to read project health at a glance and to drill into one project's task graph.

  1. Click the "Project View" tab.
  2. Each card shows the project name, status, last activity, thread count, and a plan-progress bar when a plan is running.
  3. Use the sort buttons (Recent, Name, Status) in the top-right to reorder the cards.
  4. Click a project card to open that project's task graph.

The task graph (project drill-down)

Clicking a project card opens its task graph: a canvas showing every task as a node, connected by arrows in dependency order. Colors follow the lanes; running tasks animate.

  1. Click a project card in Project View.
  2. Each task node shows its title, assigned agent, and status. Hover a node to highlight it and its connections.
  3. Drag the canvas to pan; pinch or scroll to zoom. A minimap in the corner shows the whole graph.
  4. Checkpoint tasks (ones that need your decision) appear as diamonds instead of rectangles.
  5. To return to the grid, click the back arrow or press Escape.

Project Intelligence (ask about this project, in place)

Across the top of an open project sits the Project Intelligence bar: "Ask The Architect about this project." Type a question there (why a choice was made, where things stand, what is left) and the Architect answers from that project's own record, so you never have to re-read four threads to reconstruct the story. Small lane chips next to the graph filter it by department (Build, Content, Ops), and Fit all re-frames the whole picture.

The system view showing agent workload cards and provider health
System View. Every agent's workload and current task, plus the health of each AI provider.

System View

System View is the machine-level dashboard: global stats, every agent's workload, and the health of every configured provider. It refreshes on its own every few seconds.

  1. Click the "System View" tab.
  2. If anything is blocked or erroring, a Needs attention banner appears at the top. Each item is clickable: it takes you to the blocked project or the agent's thread.
  3. The global stats bar shows active threads, project counts, blocked items, and total tokens used.
  4. The Agent Fleet section lists every agent with a status dot (green active, red error, yellow stalled), current task, and workload. Click an agent to open its busiest thread.
  5. The Provider Health section shows each AI provider, whether it is the default, and its status. A red border means an error you can fix in Settings, Providers.

The Sidebar

The Sidebar is the panel on the left of the Terminal. It lists all your threads, lets you search and filter them, and gives you access to cost tracking, files, and skills.

The sidebar with the prompt, action buttons, tab bar, filters, and thread list
The Sidebar: the "What do you need?" prompt up top, then the Sessions / Cost / Files / Skills tabs and the thread list.

The header and the "What do you need?" prompt

At the top of the Sidebar is a greeting and the "What do you need?" prompt. This is the fastest way to start anything: type a request and the app routes it to the right person or to the Architect. Two buttons sit below it: The Architect (orange) and Open Terminal (cyan).

  1. Click the "What do you need?" field (or press Cmd+N to focus it).
  2. Type your request. After a short pause, a routing chip appears below showing the suggested agent, lane, and project.
  3. If the routing looks right, press Enter to send it there. To adjust, click the chip (or the small filter icon) and pick a different agent, project, or lane.
  4. For multi-agent work ("build me X"), the request opens the Architect instead. For a quick fact or "show me," it goes to Morpheus. For a system task, it goes to the Operator.

The four tabs: Sessions, Cost, Files, Skills

  1. Sessions (default): the thread list with grouping, search, and filters. This is where you pick a thread to open.
  2. Cost: a breakdown of token spend per thread and over time.
  3. Files: the file browser, showing files agents created or that you attached.
  4. Skills: reusable skill patterns flagged from past conversations.

Grouping and searching threads

  1. Use the three-segment control above the search box to switch grouping: By Lane (grouped under colored lane headers), By Recent (one list, newest first), or By Project (grouped under each project).
  2. Type in the search box (or press Cmd+K) to filter threads by name, project, or agent. Results update as you type.
  3. Each thread row shows a status dot (green active, yellow paused, red error, gray complete), the name, the agent and project, and a time.

Good to know: Double-click a thread name to rename it. Right-click a thread for Rename and Archive. A "Show archived" toggle appears at the bottom of the list when you have archived threads.

Opening and using a thread

A thread is a single ongoing conversation with one agent. Clicking a thread replaces the Cockpit with that conversation, where you read the history, send messages, attach files, and manage the session.

A thread conversation with header, message bubbles, and the input bar
A thread. The header shows the project, agent, and model; messages stream below; the input bar is at the bottom.

Reading a thread

  1. Click any thread in the Sidebar.
  2. The header shows the project name, a status dropdown, the agent, the model, and a context meter (how full the context window is). If the project has a plan, a View Plan pill appears.
  3. Your messages align right; agent replies align left and are formatted text.
  4. Hover an agent reply for three icons: a speaker (play it aloud), a copy icon, and a flag (save the reply as a reusable skill).

Sending a message

  1. Click the input bar at the bottom and type.
  2. Press Enter to send, or Shift+Enter for a new line. You can also click the send button.
  3. While the agent responds, you see a typing indicator and the reply streams in. To stop it, click Cancel in the header.

Attaching a file, dictating, and choosing a model

  1. Attach: click the paperclip, choose a file (or folder), and it rides along with your next message. Remove it with the X on the badge.
  2. Dictate: click the microphone, speak, and your words fill the input. Click it again to stop.
  3. Model: use the model dropdown in the toolbar to override the agent's default for the next message only. It resets to "Default" after you send.

Good to know: Click "Wrap Session" in the thread header to save a clean handoff summary, so a long thread can resume later in a fresh context. If a thread hits an auth error or a rate limit, a banner appears with a "Fix Keys" or "Retry" button.

The Architect: turning a request into a plan

The Architect is the planning intelligence. You give it a brief, it can ask a few clarifying questions, then it generates a plan: a list of tasks assigned to specific agents. You review the plan, approve it, and the work dispatches automatically.

The Architect modal with a brief field, question depth, involvement, and Generate Plan
The Architect. Write the brief, set how many questions it asks and how much oversight you want, then Generate Plan.

Opening the Architect

  1. Click "The Architect" (the orange button) in the Sidebar header. The Architect opens over the Terminal. You can also reach it from a project's Orchestrate button, or by typing a build request into "What do you need?".
  2. Type your request into the brief box. Be specific: the deliverable, the scope, any constraints.
  3. Optionally attach files for context (drag them onto the attachment area).
  4. Choose a question depth (Off, Light, Default, or Aggressive): how many clarifying questions it asks first.
  5. Set the involvement level: how often agents pause for your approval, from fully supervised to full auto. Set Max Turns to cap how long each agent runs before checking in.
  6. Click Generate Plan.

Answering its questions

  1. If question depth is on, the Architect asks one or more rounds. Each question has an answer box; on desktop it reads the question aloud when you focus it.
  2. Type your answers, or click Skip on any you do not want to answer.
  3. Click Submit Answers to continue, or Generate Plan Now to stop the questions and plan with what you have given.

Reviewing and approving the plan

  1. When the Architect finishes, you land in the lead thread and the plan fills the main area: a status badge, a summary, a task count, and a progress bar.
  2. Each task is a card with its title, agent, lane, and status. While the plan is a draft, you can reorder tasks (up/down arrows), reassign an agent or lane, remove a task, or Add Task.
  3. When it looks right, click Approve. The status changes to Executing and agents begin. To discard it, click Cancel Plan.

Watching it run, and checkpoints

  1. The progress bar fills as tasks complete. Click View in Terminal to watch it as an animated task graph in the Cockpit.
  2. If a task hits a checkpoint (a decision only you should make), it turns amber and appears in a Decision Queue above the task list, and in your Briefing.
  3. Click the checkpoint's action (approve, reject, or the specific options it offers) to unblock that task and let the work continue.
  4. When everything finishes, the badge reads Complete.

The Decision Room

The Decision Room is the short list of things that are genuinely waiting on you. The system and your agents work on their own as far as they safely can; when they hit something only you should decide (a spend, a fork, a checkpoint, an out-of-scope question), it surfaces here as a card.

To work through your decisions

  1. Click the Decision Room room icon (bottom-left).
  2. Read each card. A card explains what came up, why it needs you, and what the options are.
  3. Choose the action on the card (for example approve, decline, pick an option, or open the detail). The agent waiting on that answer picks the work back up immediately.

Good to know: If the room is empty, that is the system telling you nothing needs your call right now. That is the goal, not a bug.

Why a card appears. Common reasons: a spending decision, a credential is needed, a plan reached a checkpoint you asked to confirm, a request was out of scope, an agent is genuinely blocked and needs direction, or a model you use just got a new default version.

When the default model changes

The Construct keeps a built-in registry of which models exist and which one is the default for each family. The current default is Claude Opus 4.8 (the default for the Opus family and the Frontier tier, replacing Opus 4.7). When that default moves, The Construct does not switch your agents silently: a card appears here titled like "New model available: Claude Opus 4.8," letting you know the agents that were on the previous default have been moved to the new one (same context window, generally stronger on hard reasoning). Click Review my agents to jump to Settings → Agents, or Got it to dismiss it.

Convene the Council on a hard call

For the decisions that matter most, you do not have to settle for one model's opinion. Click Convene the Council (the ⚖ button) to put the question to a panel of models that draft, critique, and synthesize a verdict together. The full walkthrough is in the next section.

The Decision Room, clear: Decisions clear, with the Convene the Council button ready
The Decision Room on a good day: "Decisions clear." The Convene the Council button waits at the top for the hard calls.

The Doctor: what the system sees

In the bottom corner of every room sits a small ◎ button. Click it (or press Cmd+Shift+O from anywhere) and The Construct shows you its own vital signs, and listens to yours.

The Doctor overlay: the System Status checklist above, the Capture Observation box below
The Doctor's two halves: System Status (the app checking its own vitals, green OK or orange Warning) and Capture Observation (tell it what you saw).

System Status (the top half)

A live health readout of the whole machine: is the AI provider reachable, are credentials valid, are any build threads stale, is the filesystem access intact, is the supervisor running. Each line shows a green OK or an orange Warning. You never have to interpret it; it is there so that when something feels off, you can see at a glance whether the system already knows.

Capture Observation (the bottom half)

  1. Click the ◎ button (or press Cmd+Shift+O).
  2. Type one honest sentence about what you just saw: "the voice preview stayed quiet," "this plan seems stuck," "I expected a button here."
  3. Press Cmd+Enter (or click Save).

Your observation is recorded durably, along with which room, project, and thread you were in, so nothing gets lost or has to be re-explained, and it surfaces as a card in your Decision Room. Separately, when a System Status check fails and the Doctor knows the fix, it proposes a remediation you can preview and approve. Complaining, as a feature: the fastest way to make the app better is to tell it what you saw, the moment you see it.

Good to know: observations are for anything worth remembering, not just problems. "This worked beautifully, do it this way again" is a perfectly good observation.

The LLM Council

Some calls are too important to leave to a single model. The LLM Council is a panel of AI models that independently draft an answer, critique each other, and elect one of their own to synthesize a final recommendation. You get a reasoned verdict with the disagreements surfaced, not one model's first guess.

Two ways to convene it

The council deliberating in the Dojo: the model voters up top, the six perspective hats below, and a Stop button on the right.
The council deliberating in the Dojo: the model voters up top, the six perspective hats below, and a Stop button on the right.

How a run works

Each run has two kinds of participants:

The chair then synthesizes everything (its own winning draft, the other voters' drafts, and the six perspectives) into one final answer, with the Adversary's strongest challenge addressed head-on.

Make the panel yours. The default lineup works out of the box, but in Settings → Council you can hand-pick the voter models (two to four), turn perspective advisors on or off (or write your own), and tune how many rounds it deliberates. Leave it untouched and it keeps choosing your strongest connected models for you.

The finished verdict, kept in the thread.
The finished verdict, kept in the thread.
You control the spend. A council run uses your own provider keys, so it costs real tokens. A Stop button sits on the panel the whole time it runs. Click it and the council halts at once, dropping any in-flight model calls so you never pay for more than you have already used. A run also stops itself if a model goes quiet for too long, so it can never hang forever.

Settings

Settings is where you configure everything: your name and theme, the AI providers, your agents and their voices, capacity, notifications, backups, and more. You can open it any time without losing your place.

The Settings panel open to the General tab
Settings opens as a full panel with the tabs down the left side.

To open Settings

  1. Click the gear (settings) icon. In the Terminal it sits at the bottom of the Sidebar.
  2. Click any tab on the left to jump to that section.
  3. Press Escape, or click the X in the top-right, to close Settings and return to your work.

Good to know: Most changes save the moment you make them. A brief "Saved" mark appears when the value is written. Anything that needs a restart says so.

Settings: General

General controls your identity, the look and feel, how much personality the app shows, how thoroughly the Architect checks its own work, and the default models and oversight used everywhere.

The General tab
General: name, personality level, theme, font size, concurrency, and default models.

What is here

New here? Start on Off or Lite. The app opens in plain-language mode, and while you are still learning where things are, Off or Lite keeps every instruction, tooltip, and button label as clear as possible. Once you know your way around, turn the personality up to Medium or There Is No Spoon for more character. You can change it any time, and nothing about how the app actually works changes with it: only the wording. (Higher levels also rename some settings into theme-flavored terms, which is one more reason to learn on a lighter setting first.)
Two model settings, and you can leave both alone to start. The Default Model is what your agents use to do the actual work. The Orchestration Model (renamed Planning Model on higher personality levels) is what the Architect uses to plan and coordinate a job before the work begins. They do not have to match, and they do not have to differ: keeping both on their defaults is the right choice until you have a specific reason to change one.

To change the theme

  1. Open Settings, then the General tab.
  2. Under Theme, click the one you want. The app switches immediately.
  3. If you pick Matrix below the top personality level, a small prompt offers to raise the personality to match. Choose Yes or No.

Settings: Providers

Providers are the AI services that power your agents. Connect as many as you like, mark one as the default, and test each connection without leaving the app.

The Providers tab
Each provider is a card with Test, Edit, Set as Default, and Delete. The default has a blue badge.
Read this first: subscription vs. API, and most people want the subscription. There are two ways to power The Construct. If you already pay for Claude Pro or Max, a ChatGPT plan, Google Gemini, or GitHub Copilot, you can just use that: no extra signup and no separate bill. An "API key" is a different thing: a separate, pay-as-you-go developer account, billed apart from your monthly plan. The Type names below tell you which is which, so you do not have to guess.

The provider types

Use a subscription you already have. No separate bill; it connects through that service's own app or command-line tool:

Use an API key. A separate developer account with its own pay-as-you-go billing (see "Getting your API key" below):

OpenClaw: the local OpenClaw gateway on your Mac (default http://127.0.0.1:18789), if you use it.

How to tell at a glance: if the Type name says CLI, plan, or names a subscription like Copilot, it uses what you already pay for. If it says API key, or asks you to paste a key, it is the separate pay-as-you-go path.

To add a provider

  1. Open Settings, then Providers, then click Add Provider.
  2. Pick the Type, give it a Display Name, and enter the Base URL and API Key if the type needs them (click "Show" to see the key as you paste).
  3. Click Save. The provider appears in the list.
  4. Click Test to confirm it connects. Green dot and a model count mean success; a red dot shows the error.
  5. Click Set as Default on any provider to make it the one agents use by default.

Good to know: When you Edit a provider, leave the API key blank to keep the existing one.

Getting your API key (start here if you are new to this)

An API key is what lets The Construct talk to an AI on your behalf. One thing trips up almost everyone at first: an API key is not the same as your $20-a-month ChatGPT or Claude subscription. The API is a separate signup with its own billing, usually pay-as-you-go. You make a free account on the provider's developer site, create a key, and add a little credit (a few dollars goes a long way).

Where to get one:

Copy the key the moment it appears (you usually cannot see it again), then paste it into The Construct.

If a screen looks different from these steps, ask your AI. These developer sites rearrange their buttons often, so printed steps go stale fast. The quickest way to unstick yourself is to open ChatGPT, Claude, or Gemini and ask, in plain words: "How do I create an API key for [Google Gemini, Anthropic, or OpenAI] right now?" It knows today's screens and will walk you through them, one step at a time.

Settings: Agents

Agents are the people of The Construct. Each has a name, a role, a model, a lane, and optional custom instructions. You can build them from a template, edit any of them, or reset the whole roster.

The Agents tab
The roster. Each agent shows its display name, lane color, and description, with an Edit button.

What each agent has

About Provider and Model, in plain terms: you rarely need to change these. Every agent runs on whichever Provider you connected, using a sensible Model by default. If you do want to change them, set the Provider first (which AI service this agent should use); the Model list then shows that provider's own models, so the two can never disagree. Leaving Provider on "Default" simply follows the default provider you set on the Providers tab.

To add an agent from a template

  1. Open Settings, then Agents, then click Add Agent.
  2. Pick a template (Blank, Trinity, Research Analyst, Data Engineer, Security Auditor, or Social Media). The form opens pre-filled.
  3. Set the Display Name, Description, Lane, and Model, add a System Prompt if you like, then click Save.

Good to know: To edit, click Edit on any agent (the Name is locked after creation). "Reset Defaults" restores the standard roster but does not erase the voices you set in Personalities, which are stored separately.

Settings: Personalities

Personalities sets a display name and a speaking voice for each agent. The name shows everywhere the agent appears; the voice is used whenever the agent speaks aloud.

The Personalities tab
Each agent row: an editable name, a voice dropdown, and a Talk button to start a live voice chat.

To rename an agent or give it a voice

  1. Open Settings, then Personalities.
  2. To rename: click the agent's name field, type the new name, and press Enter. It updates everywhere.
  3. To set a voice: click the voice dropdown and choose one (Brian, Eric, George, Sarah, Alice, and others, or Default to use the global voice). A short sample plays so you can hear it.
  4. To talk live with that person: click the Talk button. The voice door opens addressed to them.

Good to know: A voice set here overrides the global voice for that one agent. This is the tab where you choose who answers your phone, too.

Settings: Lanes

Lanes organize your work into named, colored groups (such as Build, System, Content, and Ops). The color identifies the lane in agent badges and the sidebar. You can add lanes, drag to reorder, and reset to defaults.

The Lanes tab
Lanes in display order, each with a color swatch and an Edit button. Drag a row to reorder.

To add or reorder a lane

  1. Open Settings, then Lanes.
  2. To add: click Add Lane, enter a Name, a Display Name, pick a Color, add an optional Description, and click Save.
  3. To reorder: point at the drag handle on the left of a row, drag it up or down, and release. The order saves automatically.
  4. To change one: click Edit, adjust the Display Name, Color, or Description, and Save.

Good to know: Deleting a lane does not delete its agents; they just show a missing-lane color until you reassign them.

Settings: Capacity

Capacity controls how many agent sessions run at once and what happens when the limit is reached. Two presets cover most needs; a Custom mode gives you every dial.

The Capacity tab
Pick Lean or Unleashed, or click Customize for the detailed controls.

The presets

To configure capacity in detail

  1. Open Settings, then Capacity, then click a preset or Customize.
  2. Drag Max concurrent sessions to your limit.
  3. Toggle Prefer top model on or off.
  4. If you want overflow to go to a paid provider, toggle Allow spillover on, choose the spillover provider, and set a monthly budget cap.
  5. Choose what happens at the cap: Pause (queue), Fallback (use the spillover provider), or Error (stop).

Settings: Projects

Projects group related threads under one name, set a lead agent, and set how much autonomy that project's agents have. Create projects, assign threads to them, and set their guardrails here.

The Projects tab
Each project card shows its status, control level, lead agent, and thread count. Expand it to manage threads.

What a project has

To create a project and assign threads

  1. Open Settings, then Projects, then click Add Project.
  2. Enter a Display Name (the slug fills in automatically), optionally pick a Lead Agent and add a Description, and click Save.
  3. To attach threads: click Expand on the project, pick an unassigned thread and a role (primary, supporting, or review-gate), and click Assign Thread.

Good to know: Deleting a project removes its thread assignments but not the threads themselves.

Settings: Voice

Voice controls both how you speak to The Construct (on-device speech recognition) and how it speaks back, plus speed, timing, and muting. Out of the box the app speaks with your Mac's built-in voice: free, private, and no key to enter.

The Voice tab
Two parts: Voice Input (speech to text) and Voice Output (text to speech).

Voice Input (speaking to the app)

The Construct recognizes your speech with an on-device model, so your audio never leaves your Mac. If it is not installed, click Download voice model (about 148 MB, one time), wait for it, then restart the app.

Voice Output (the app speaking)

Good to know: If voice is not responding, check that your Personality Level (General tab) is Medium or higher, which is required for full spoken back-and-forth.

Settings: Notifications

Notifications controls when and how the app alerts you: whether they are on, which events trigger them, how often, and whether they make a sound.

The Notifications tab
Turn notifications on, then tune exactly which events you want to hear about.

To turn notifications on and tune them

  1. Open Settings, then Notifications, then click Enable Notifications. Allow the permission when your system asks.
  2. Click Send Test to confirm one arrives.
  3. Set the Frequency: Every event, Batched, or Important only. Toggle Sound on or off.
  4. In the event checklist, turn individual events on or off (session complete, session error, context critical, agent needs input, gateway down, and more).

Good to know: If the Enable button is greyed out, your system has blocked notifications. Allow them for The Construct in your system settings, then come back.

Settings: Backups

Backups lets you take a snapshot of The Construct's data at any time and restore one later. It keeps up to ten, pruning the oldest automatically.

The Backups tab
Create a backup with one click; each saved snapshot shows its date, time, and size, with a Restore button.

To back up and restore

  1. Open Settings, then Backups.
  2. Click Create Backup. When it finishes, the new snapshot appears at the top of the list with its size.
  3. To restore: find the snapshot, click Restore, confirm the warning, and restart the app when it is done.

Good to know: Take a backup before big changes (resetting agents, restructuring projects). It is a one-second undo path.

Settings: Access

Access is where the app becomes yours: buy a lifetime license, or enter a license key or beta code. One field takes either.

The Access tab
The Access tab (called Beta before 0.1.49): buy button, one redeem field, and the advanced toggles below.

The three ways in

The advanced toggles

Good to know: Agents that can write to shared memory can add new notes but cannot overwrite or delete notes another session wrote. The shared memory is treated as an append-only record.

The Playbook: Start the Adventure

The guide tells you where everything is. The Playbook makes you use it: seven quests, five to twenty minutes each, every one built around something real from your own life, with briefs you can copy and paste exactly as written.

It lives at theconstructapp.com/playbook: read it online, or keep a copy the same way as this guide (File, then Print, then Save as PDF). Like everything here, every step in it was played through on a real copy of the app before it was written down.

Your phone: text The Construct over Telegram

The Construct can answer your texts. Pair your phone once, and a message you send from anywhere reaches your chosen agent on your Mac; the reply comes back in the same chat. The person who answers is the Concierge: warm, brief, phone-friendly. Nothing new is exposed to the internet; your Mac reaches out to Telegram to collect messages, the way a browser fetches a page.

Set it up (about five minutes, all in Settings → Messaging)

  1. Create your bot. In the Telegram app on your phone, message @BotFather, send /newbot, and follow its two prompts. It hands you a token. This bot is your private line; only chats you pair can use it.
  2. Paste the token into Settings → Messaging → Bot token. It is stored in your Mac's Keychain, never in the app database. The panel shows Connected as @your-bot when the token works.
  3. Pair your phone. Click Generate pairing code, then send that code to your bot from your phone. The code works once; the panel then lists the paired chat. Anyone else who finds your bot gets silence.
  4. Turn the switch on (top-right of the Telegram card) and click Send a test message to confirm the line works.

What you can do from your phone

Set who answers. The agent that replies to your phone is chosen in Settings → Messaging (the Who answers dropdown; the Concierge by default), not in Personalities.

iMessage answering (for a dedicated-number setup) is configured separately and is off unless you turn it on; for most people, Telegram above is the ready path.

The safety line

The Construct has two modes, and they sit on opposite sides of this line. Knowing which one you are in is the whole game.

Talking with Morpheus, in the Dojo or by voice

This side is deliberately hands-off. Morpheus will open pages, browse, search, fetch live facts, read, draft, and lay out options, but it is instructed to stop short of spending money, entering a credential, submitting a form, buying, booking, or anything it cannot undo, and to hand that last step to you. Treat that as a sensible default for casual back-and-forth, not an unbreakable lock: it is a standing instruction to Morpheus, not a wall in the code. (Work you hand off in conversation, "Tank, build...", goes to your agents, which run under the gates below.)

Running agents, through the Architect and your team

This side is built to execute. When you approve a plan, or run a task at Full Autonomy, your agents use the tools and credentials you have given them to carry the work out, including steps that cost money or cannot be undone. That is the point of having a team. The safeguards here are the ones you set, and they are real:

The honest version: the more autonomy, tools, and credentials you grant, the more The Construct will do on its own, including things that spend money or cannot be undone. It does not secretly hold itself back. You decide how much rope to give it, up front, with the gates above.