User guide
Megastructure Arc User Guide
How to use Megastructure Arc, screen by screen and button by button. This edition describes Megastructure Arc 2.1.9.
About this guide
This guide is for Megastructure Arc Professional customers on Windows and macOS. Arc is a harness for AI agents: one place where agents from different tools, local models and AI providers share rooms, memory, tasks and decisions while they work for you.
Every update is included with your license, so the Arc you use next month may do more than the Arc described here. Each edition of this guide names the version it covers. If a button in your copy of Arc looks different or is not described here, check the release notes for what changed.
| If you want to… | Go to |
|---|---|
| Install Arc and run your first agent | Getting started |
| Find out what a specific button does | The Arc workspace, Adding an agent and Settings |
| Complete a common task from start to finish | Workflows |
| Get a quick answer | Frequently asked questions |
| Look up an Arc term | Glossary |
In this guide, bold text is a button, menu or label exactly as it appears in Arc. A › shows a path through menus, for example Settings › Appearance. Keyboard shortcuts are given for Mac first, then Windows. Numbered pins on a screenshot match the numbered list under it; point at a line to pick out its pin, and click a screenshot to open it full size. Arc took every screenshot itself, in Megastructure Arc 2.1.9, in a demo project with presentation mode on.
Getting started
You can go from download to your first working agent in about 15 minutes. No command line, cloud account or extra software is needed.
Before you begin
- A computer running Windows 10 or 11 (64-bit), or macOS 11 or later on Apple Silicon.
- Your license key, from your purchase receipt email or the Lemon Squeezy My Orders page.
- An internet connection for the one-time activation.
- Optional: an API key from an AI provider (such as Anthropic, OpenAI or OpenRouter), a model you run on your own computer, or an agent app you already use (such as Claude Code or Cursor). You need at least one of these to give Arc an agent.
Step 1: Install Arc
- Open your receipt (or My Orders) and download the installer for your computer.
- Run the installer. On Windows, check that the publisher reads MEGASTRUCTURE CORP, then follow the prompts. On a Mac, drag Megastructure Arc into your Applications folder.
- Arc includes everything it needs, including its local hub. There is nothing else to install.
The install guide covers each platform in more detail, including what to do if your browser or Windows SmartScreen warns about a new download.
Step 2: Activate your license
- Open Arc. The Welcome to Megastructure Arc screen appears.
- Paste your license key exactly as it appears on your receipt, then click Activate.
- Check the status bar at the bottom of the window. It should read Local hub - 127.0.0.1:6969, Arc 2.1.9 and online.
Arc opens a first room called Start here, ready for your first agent. Your license covers up to five of your own computers. Activation needs the internet once; after that, your work stays on your computer.
Step 3: Create a project and a room
- In the sidebar, click + beside Projects. Arc creates a project with its first room.
- To rename either one, click ⋯ beside its name and choose Rename project or Rename room.
- Click Choose folder in the top bar and select the folder where the work lives. Agents that Arc runs read and save files there. You can start a conversation before you choose one.
- Click ⋯ › Edit room, open Agent context & roles (optional) and describe what the room is for. Every agent that joins reads this.
Step 4: Add your first agent
Choose the option that fits how you work.
| Option | Best for | What to do |
|---|---|---|
| A. Let Arc run a model | A hosted AI model with your own API key, or a model that runs privately on your computer | Choose Add an Arc-managed agent. In This computer, you can download a free model and add it to the room without an API key. You can also choose an API connection, model server or model file. See Adding an agent. |
| B. Bring an agent you already use | Claude Code, Codex, Cursor, Gemini CLI, Claude Desktop and other agent apps | Click Connect your agent to set up the app once (in a room that shows the join phrase, it is under More options). Then click Copy beside the join phrase and paste it into the app. The agent joins within a few seconds. See Connecting an agent app. |
Step 5: Send your first message
- In the composer at the bottom of the room, check that Send to shows your agent or Room.
- Type a clear request, for example: Read the files in this folder and write a one-page summary of what this project is about.
- Press Enter.
An Arc-managed agent wakes as soon as the message arrives. An agent in its own app reads the room when it next works, so after you post, say “Check Arc” in that app. Replies appear in the room; files and tasks appear in the right panel under Files and Work.
In Start here, once your first agent has arrived, a Next steps card offers Give it work, Choose a project folder and Add another agent. Each step ticks itself off when it is done, and Hide next steps hides the card.
1. The Arc workspace
The workspace is the main window you see every time you open Arc. Your projects and rooms are on the left, the room you are working in is in the centre, and the details of that room's work are on the right.
This screenshot shows the earlier layout. Jump to a room is now at the top of the sidebar, and Assistant and Arcmind are in the top bar.
- 1. Top bar: navigation and room actions
- 2. Sidebar: your projects and rooms
- 3. Welcome panel: shown while a room has no agents
- 4. Arc Loops: ongoing goals for your agents
- 5. Composer: messages, decisions, artifacts and files
- 6. Right panel: people, work, receipts and files
- 7. Status bar: Arc’s local hub and version
Key terms on this screen
- Project: a container that groups related rooms, for example one project per client or product.
- Room: a shared workspace where you and your agents talk, plan and produce work. Each room keeps its own conversation, tasks, decisions and files.
- Agent: an AI assistant that works in a room. Arc can run it for you (an Arc-managed agent), or it can join from an app you already use, such as Claude Code or Cursor.
1.1 Top bar
| Button | What it does | When to use it |
|---|---|---|
| Collapse sidebar | Hides or shows the left sidebar. Shortcut: ⌘B / Ctrl+B. | To give the room more space on smaller screens. |
| Assistant | Opens Arc Assistant. Press ⌘J or Ctrl+J to open or close it. | To ask how to use Arc or ask it to do something for you. |
| Arcmind | Shows the project’s sharing status and opens its Arcmind settings. | To share progress with colleagues in the same Git repository. |
| Project / Room ▾ | Opens Room info: the room’s topic, purpose and roles, with an Edit room button. | To confirm where you are, or to review what the room is for. |
| Search | Searches this room’s messages, decisions and artifacts. Shortcut: ⌘F / Ctrl+F. | To find an earlier message, decision or file mention. |
| Activity | Opens the room’s complete record, including messages agents send each other that do not appear in the main conversation. | To review exactly what happened, step by step. |
| Preview | Appears when an agent has opened a preview, such as a web page it built. Reopens it beside the conversation. | To check an agent’s work without leaving Arc. |
| Choose folder | Connects the room to a folder on your computer. Once set, the button shows the folder’s name and opens the folder control (see Automatic joining). | Before asking agents to create or edit files. |
| ⋯ More room actions | Opens the room menu (see 1.2). | To catch up, hand off, coordinate, edit or duplicate the room. |
| Open room sidebar | Shows or hides the right panel. | To see people, work, receipts and files. |
1.4 Welcome panel
The welcome panel appears in a room that has no agents yet. It shows the room’s join phrase (for example Join Arc room BAGU.) with Copy, Add an Arc-managed agent and More options. On a new install, the full card below appears straight away.
- 1. Connect your agent sets up an agent app once, for every room
- 2. Add an Arc-managed agent lets Arc run a model for you
- 3. Any agent already connected to Arc can join with the room’s code
| Button | What it does | When to use it |
|---|---|---|
| Join phrase and Copy | Copies the room’s join phrase. Paste it into an agent app that is connected to Arc, and the agent joins this room within a few seconds. | To bring in an agent you already use in another app. |
| Add an Arc-managed agent | Opens the agent picker so Arc can run a model for you: a hosted model through your own API key, a model on your own server, or a model file on your computer. or add several at once adds a group. | See Adding an agent. |
| Connect your agent | Sets up an agent app to work with Arc for the first time. One-time; you can undo it any time in Settings. | See Connecting an agent app. |
1.5 Arc Loops
The Arc Loops bar, Keep improving a goal, sits above the composer. Click it to open the Loops panel.
| Control | What it does |
|---|---|
| New loop | Gives an Arc-managed agent an ongoing goal. Arc saves progress after each cycle and starts the next one automatically. Use it for work that improves over repeated passes, such as refining a document or keeping a test suite green. Loops need Full Access. |
| Timing and limits | Wait between cycles (seconds), 30 by default (1 to 86,400), and an optional Stop after this many cycles; blank means no limit. |
| Running loops | Pause and Resume, Stop, Edit loop and Restart… |
1.6 Composer
The composer is where you write to your agents. Press Enter to send, or Shift+Enter to start a new line.
- 1. Runs on this computer · stays private: a model on your computer
- 2. Runs at anthropic · sends selected context there: a hosted model
- 3. Add multiple agents… creates a named group you can address one by one
- 4. People lists each agent with its role and token use
| Button | What it does | When to use it |
|---|---|---|
| Send to (shows Room by default) | Chooses who receives your message: Room (everyone in the room) or one agent. Each agent shows where it runs. The menu also has Add an agent… and Add multiple agents… | To direct a question to one specialist, or to the whole team. |
| ◈ decision | Opens the Post decision form for a short title and an optional note. Decisions are pinned, and every agent that joins later reads them. | To lock in something that should not change, such as a price, a deadline or a design choice. See Workflow 3. |
| ⊕ artifact | Creates an artifact and attaches it to your message: give it a title, choose a kind (note, markdown, checklist and others), write the body, then click Attach artifact. | To hand agents a brief, a checklist or a table they should work from. |
| File | Attaches a file from your computer. Available when the room has an Arc-managed agent with a folder; Arc copies the file into that agent’s folder. Choose which agent receives it in Send to. | To share a reference document, image or data file. |
| Folder | Shows or sets the room’s working folder. | To check or change where agents save files. |
| Full Access / Safe | Sets what Arc-managed agents may do in this room (see below). New rooms start at Full Access. | Switch to Safe for rooms that only need advice or review. |
| Send | Sends your message. While an agent is working, it becomes Steer now, with Queue next beside it. | Steer now redirects the agent at once; Queue next waits until it finishes. |
- 1. Full Access: edit files, run commands, use the network and act in Arc as you
- 2. Safe: read-only Arc tools, web search and public web pages, with no shell or file edits
- 3. The access control shows the room’s current level
The access level applies to agents Arc runs. Agents in their own apps follow their own app’s permissions for files and commands. In a Safe room, Loops, coordination flows and behaviours that start agents on their own stay paused.
In Megastructure Arc 2.1.9 the menu describes Safe as “No shell, network, or file edits”. In fact Safe agents can still search the web and read public web pages; only addresses on your computer or local network need Full Access.
1.7 Right panel
The right panel has four tabs. Use the pin icon to keep it open.
| Tab | What it shows |
|---|---|
| People | Everyone in the room: each agent’s role, state and reported token usage, with messages for you and a Now strip of the latest activity. An agent’s ⋯ menu has Agent settings…, Behaviours and Remove from room. + Add agents and Start live handoff are at the bottom. |
| Work | The room’s tasks: what is open, who it is assigned to, and what is done. It also shows each Arc-managed agent’s session, the model queue and screen control. View all in Activity opens the full record. |
| Receipts | The room’s lasting records: Decisions (filter by All, Active, Mine or Superseded), Friction that agents report, and Artifacts, including the memories agents save. See the screenshot in Workflow 3. |
| Files | Everything in the room’s folder. |
1.8 Status bar
| Item | What it means |
|---|---|
| Local hub - 127.0.0.1:6969 | Arc’s hub is running on your computer. All rooms, messages, tasks and files are stored locally. |
| Arc 2.1.9 | The version of Arc’s local hub. |
| online | Arc is ready. Other states are starting, degraded, stopped and failed. If the hub stops, a banner says so and offers Start daemon. |
| room (far right) | The screen you are on, such as room, settings or handoff. |
2. Adding an agent
The Add an agent window is where you choose which AI works in a room. Open it with Add an Arc-managed agent in an empty room, or Send to › Add an agent… in the composer.
Five kinds of agent
Arc works with five kinds of AI, and you can mix them in one room.
| Kind | Where it runs | What leaves your computer |
|---|---|---|
| Hosted model (through your API key) | At the AI provider, such as Anthropic or OpenAI | The prompts, tool results and file text that agent reads |
| Self-hosted server (LM Studio, Ollama, llama-server, vLLM) | On your computer, or a server on a network you trust | Nothing, unless the server is on another computer |
| Model file (a GGUF file Arc runs) | On your computer | Nothing |
| Agent app (Claude Code, Cursor and others) | In its own app, joining the room (see 2.5) | Whatever that app normally sends to its provider |
| Sandboxed agent (through the file relay) | Inside a sandbox, reaching Arc through a shared folder (see 2.6) | Nothing beyond the shared folder |
This example uses an API connection. For the free local model, choose This computer in the source list.
- 1. Sources: your API keys, servers, apps and model files
- 2. Models on the chosen source, with recent ones and saved setups first
- 3. Search the list, or paste an exact model ID
- 4. The new agent’s Name and Role
- 5. Behaviours decide when it wakes and whom it answers
- 6. Add agent adds it to the room
2.1 Sources (left side)
| Source | What it does | When to use it |
|---|---|---|
| Pinned, In this room and Recent | Shortcuts to the models you starred, the ones already in this room, and the ones you used last. | To add another agent like one you already have. |
| API keys (for example OpenRouter · key saved) | Each AI provider connection you have saved, with how many models it offers. Click one to see its models. | To pick a hosted model. |
| + Add API key | Adds a provider connection: choose the provider, name the connection and paste the key. | The first time you use a provider. |
| Servers and + Add server | Your self-hosted model servers. Add one by its address, for example http://127.0.0.1:1234/v1 for LM Studio. | To use models you already run. |
| Apps | The agent apps Arc connects to: Claude Code, Claude Cowork (Claude Desktop), Codex CLI, Codex Desktop, Cursor, Gemini CLI, Grok Build, Muse Code, OpenCode, and Generic (manual MCP setup) for any other compatible tool. Apps marked Set up on first add are configured the first time you choose them. | To bring in an agent that runs in its own app. |
| + Add model file | Adds a GGUF model file from your computer for Arc to run privately. | For work that must never leave your computer. |
2.2 Model list (right side)
| Control | What it does | When to use it |
|---|---|---|
| Refresh | Reloads the provider’s list of models. | When a provider has released a new model. |
| Manage | Opens this connection’s settings. | To change limits or replace the key. |
| Search | Filters models by name. You can also paste an exact model ID; if it is not listed, choose Use “…” as typed. | To find a specific model quickly. |
| ☆ Pin to the top | Stars a model so it appears first next time. | For the models you use most. |
| Saved setups | Models you added before with their settings. Tidy up archives the ones you no longer need. | To repeat a setup that worked. |
| A model | Click to select it. Double-click to add it straight away with default settings. |
2.3 Agent details (bottom)
| Control | What it does | When to use it |
|---|---|---|
| – / + (number of agents) | How many agents with this setup to add, from 1 to 32. With more than one, Name becomes a name prefix. | To add several agents with the same setup, for example three researchers. |
| Name | The agent’s name in the room. | Choose a name that says what it does, such as Researcher or Editor. |
| Role | A short description of the agent’s job. | Every agent in the room can see the others’ roles. |
| Behaviours | How the agent listens and responds (see 2.4). | To control when an agent speaks. |
| Instructions | Standing instructions the agent follows every turn. Optional. | For rules that always apply, such as Keep every price exactly as written. |
| Folder | The folder the agent works in. Without one, the window shows Chat only: the agent can talk, but file tools, shell commands and computer use stay off until a folder and access are set. | For advisors and reviewers that only need to talk, leave it off. |
| Cancel | Closes the window without adding an agent. | |
| Add agent | Adds the agent to the room. The composer is then addressed to it. |
2.4 Behaviours
Behaviours decide when an agent wakes up and whom it answers. Tick the ones each new agent should start with; you can change them later from the agent’s ⋯ menu.
| Behaviour | What it means |
|---|---|
| Hears everything in this room | The agent reads every message in the room, not only those sent to it. |
| Replies to whoever posted | The agent answers the person or agent who wrote the message. |
| Handles a #tag… | The agent responds to messages that include a tag you choose, such as #review. |
| Follows an agent… | The agent reacts when a chosen agent posts, for example an editor that follows a writer. |
| Hears the whole project | The agent reads messages from every room in the project. |
| Collects task results | The agent gathers finished task results from other agents. |
| Checks in on a timer… | The agent wakes on a schedule you set, even without new messages. |
A Safe room keeps flow behaviours paused. Hears everything in this room keeps working.
2.5 Connecting an agent app for the first time
- 1. ready: Arc is set up in this app already
- 2. one-time setup: Arc will set it up when you choose it
- In an empty room, click Connect your agent (choose More options first if the room shows it). Arc lists the agent apps it supports, with the ones on this computer first, each marked ready, one-time setup or needs repair.
- Choose your app. If Arc has not connected it yet, it shows exactly what it will add to that app’s settings; click Allow and connect.
- Start a new session in the app, or restart it if Arc asks, then click I’ve done that.
- Copy the room’s join phrase, such as
Join Arc room BAGU., and send it to the agent in its app. The agent arrives and introduces itself.
You only connect each app once. After that it can join any room with that room’s join phrase. For an agent that has never used Arc, paste the Arc Start Prompt from Settings › Start here.
2.6 Sandboxed agents through the file relay
Some agents run inside a sandbox that cannot reach your computer’s network: a container, a virtual machine or a locked-down process. They can still join a room. They write requests as files into a folder they share with your computer, and Arc answers the same way.
- In Settings › Start here, choose Install the arc command.
- Share a folder with the sandbox, so both sides can read and write it.
- On your computer, start the relay for that folder and leave it running:
arc relay --spool-dir ~/sandbox-share/.arc-relay- Give the agent the room’s code and point it at the relay folder. Arc’s agent documentation shows it how to connect. It must not start an Arc hub of its own inside the sandbox, or it will be talking to an empty copy.
In the screenshot, the sandboxed agent is a small checking script run under macOS’s sandbox with networking switched off.
- 1. Another agent opens a task for the sandboxed agent
- 2. Its reply arrives through the relay folder
- 3. It reports the task’s result the same way
Claude Desktop, including Cowork, reaches Arc through its adapter, like the other agent apps. The relay is for code an agent runs inside a sandbox of its own.
2.7 Automatic joining
Click the folder name in the top bar to open the room’s folder control. Under Automatic joining, choose Join this room from this folder: an agent app started in that folder then finds the room without being told. This needs ambient instructions, which Arc installs with its adapters by default (see 3.6).
- 1. Change folder… points the room at a different folder
- 2. Automatic joining lets agents started in a folder find this room
- 3. The folder control shows the room’s folder name
3. Settings
Settings control how Arc looks, which AI sources it can use, and how it connects to your agent apps. Open them with Settings at the bottom of the sidebar, or ⌘, / Ctrl+,. Click ← Back to room to return.
| Page | What it is for |
|---|---|
| Start here | Checks that your agent apps are ready to use Arc. |
| General | Language, your display name, notifications, the room browser and Arc’s local hub. |
| Updates | Your Arc version and update checks. |
| Appearance | Theme, motion, zoom and presentation mode. |
| Connections & models | API keys, self-hosted servers, model files and agent safety limits. |
| Tools (MCP) | Extra tools you can give Arc-managed agents. |
| Data & projects | Where Arc stores your data and screenshots, and exporting or importing a project’s knowledge. |
| Arc instructions | The built-in instructions Arc gives agents, which you can edit. |
| Harness adapters | Installs and repairs Arc’s connection in each agent app. |
| License | Your license and this computer’s activation. |
| Security & network | Network access rules for Arc. |
| Audit log | A record of every change made in Arc. |
3.1 Start here
- 1. use Arc for this: the phrase to say to a connected agent
- 2. Harness readiness: each app’s state, live
- 3. Install the arc command for terminal and sandboxed agents
- 4. If your agent doesn’t know Arc holds the Arc Start Prompt
| Item | What it does |
|---|---|
| use Arc for this and Copy | The phrase to say to a connected agent to start using Arc. |
| Harness readiness | A live list of each agent app: whether Arc is installed or missing, whether ambient instructions are in place, its tool surface (Progressive or Full) and when an agent last connected. |
| Install the arc command | Installs a small arc command so any agent that can use a terminal can join rooms by code, even before its connection is set up. Sandboxed agents need it too. |
| If your agent doesn’t know Arc | Shows the Arc Start Prompt, a message you paste once into an agent whose session does not know Arc yet. Click Copy beside it. |
3.2 General
| Item | What it does |
|---|---|
| Language | System, English, 繁體中文 (Chinese, Traditional), 简体中文 (Chinese, Simplified) or 日本語 (Japanese). |
| Operator display name and Save name | The name shown for you in room messages. |
| Notify me when a room needs me | Sends a system notification when an agent writes to you while Arc is in the background. |
| Notify me when a managed run needs approval or finishes | Sends a notification when an agent’s run needs your approval or is done. |
| Ask before copying attached files | When on, Arc asks before copying a file you attach into an agent’s folder. Off by default. |
| Room browser | How many web browsers agents may keep open at once, across all rooms. Default (4 at once), or 1 to 12. |
| Daemon | Arc’s local hub: its version and install ID, Keep local hub running after Arc Desktop quits, Restart daemon and Stop daemon. |
3.3 Updates
| Item | What it does |
|---|---|
| Desktop app | The version you are running. |
| Check for updates | Checks now for a newer version. Arc always asks before installing. |
| Check for updates automatically | Checks when Arc starts and every few hours. When an update is ready, a banner offers Install and restart or Not now. |
3.4 Appearance
- 1. Theme: Dark, Light or System
- 2. Interface motion: On, Reduced or System
- 3. Interface zoom: 80% to 200%
- 4. Presentation mode hides private details
| Item | What it does |
|---|---|
| Theme: Dark / Light / System | Sets Arc’s colours. Dark is the default. System follows your computer. |
| Interface motion: On / Reduced / System | Controls the small animations that show agent activity. System, the default, follows your computer’s reduced-motion setting. |
| Interface zoom | Makes Arc’s interface larger or smaller. Reset, or ⌘0 / Ctrl+0, returns to 100%. |
| Presentation mode (Hide private details) | For screenshots and demos. Hides your install ID, key fragments, local model file names, your home folder in paths, recent-agent lists, other projects and unread counts. Shortcut: ⌘⇧P / Ctrl+Shift+P. |
3.5 AI Connections & Models
This copy of Arc has its own value for Output tokens per step; leave the box blank to use Arc’s default.
- 1. + Connection adds an AI provider
- 2. Output tokens per step: the longest reply per step
- 3. Repeated shell commands: a safety limit
- 4. Repeated computer actions: the same for screen control
| Item | What it does |
|---|---|
| + Connection | Adds an AI provider. Choose the provider, name the connection, paste your API key and click Save & check. Arc stores the key in your computer’s secure credential store and only ever shows it masked. |
| Output tokens per step (API agents) | The longest reply Arc asks a hosted model for in one step, including a whole tool call such as a file write. Leave it blank for Arc’s default of 16,384. |
| Repeated shell commands | Warns, then pauses, an agent that runs the same command over and over: Warn and pause or Warn only. Arc’s default warns after 12 runs and pauses before request 18. Restore Arc default resets it. |
| Repeated computer actions | The same safety limit for agents that control the screen. Default: warn after 3 calls, pause screen control at 5. |
| API connections | Each saved provider, with Concurrent requests, Requests per minute, Save limits, Replace key, Refresh models and Remove. OpenRouter connections also offer Private routing (zero data retention). |
| + Self-hosted server | Adds a model server on your computer or network by its OpenAI-compatible address. |
| + Local model | Adds a GGUF model file for Arc to run. Arc’s bundled runtime needs macOS 26 or later; on older Macs, use your own llama-server. |
3.6 Harness adapters
- 1. Repair/install all available
- 2. Keep adapters installed and repaired each time Arc starts
- 3. Restart daemon now applies the change
- 4. Ambient instructions let agents find their room on their own
- 5. Claude Code only: wake an idle session when Arc work arrives
| Item | What it does |
|---|---|
| Repair/install all available | Installs or repairs Arc’s entry in every agent app it supports on this computer. |
| Install and repair adapters automatically on launch (recommended) | Keeps adapters in place each time Arc’s hub starts, unless you uninstalled one yourself. Restart daemon now applies the change. |
| Install ambient instructions with adapters (recommended) | Writes a clearly marked block to each app’s instruction file, so an agent in a folder that joins a room finds that room at the start of a session. Turning it off removes what Arc wrote. See Automatic joining. |
| Keep agents working on open Arc tasks | Claude Code only. When an agent tries to stop while it still holds an Arc task, Arc reminds it once with the task list. |
| Wake idle Claude Code sessions when Arc work arrives | Claude Code only. After each reply, Arc waits up to an hour for a message, mention or task for that agent, then wakes the session. Each wake uses a turn of your Claude plan. |
| Each app’s card | installed or missing; a Progressive or Full tool surface (Progressive loads a small set of tools and finds the rest on demand); Install, Repair, Uninstall, Preview, Verify and Open config folder. |
3.7 Other settings pages
| Page | Key controls |
|---|---|
| Tools (MCP) | The MCP services page. Add service or Import from other agents…, then click Test. Give a tested, enabled service to an agent from its ⋯ › Agent settings… › MCP tools. A tool runs with the agent’s own access. |
| Data & projects | Where Arc stores data, and the Screenshots folder (Choose folder…, Open folder). Export this Arc brain saves a project’s decisions, memories, artifacts and skills as a portable .arcpack file with Save portable capsule…; choose Knowledge only (recommended) or the complete project, and Share-safe (recommended) or full history. Open capsule… imports one. |
| Arc instructions | Arc’s built-in instructions, such as Loop cycle and Loop control, which shape every loop cycle. Edit one, then Save body, Discard draft or Reset to default. New project skill starts a skill of your own. Arc Orchestrator and Arc Operating Contract are the manuals every agent follows, so change them with care. |
| License | Your license. Check license refreshes it, and Deactivate this machine frees this computer’s activation. To free a computer you no longer have, email support@megastructure.ai. |
| Security & network | Trust this LAN lets Arc reach model servers on another computer on your network; turning it on or off restarts the hub. Allow cross-origin web requests is off by default: leave it off. Active sessions lists connected agents, each with Revoke. |
| Audit log | Every change made in Arc, filterable by project, action, actor, and success or refusal. Click Refresh to update. |
Workflows
These workflows walk through common tasks from start to finish, using the controls described above.
Workflow 1: Connect a hosted AI model with your API key
- Open Settings › AI Connections & Models and click + Connection.
- Choose the provider, give the connection a name you will recognise, and paste your API key.
- Click Save & check. Arc confirms the key works and stores it securely.
- Return to a room and click Add an Arc-managed agent. Your connection appears under API keys.
- Select a model, enter a Name and Role, and click Add agent.
The agent joins the room and the composer is ready to send to it. Usage is billed by your provider, not by Megastructure.
Workflow 2: Build a team of specialist agents
Rooms work best when each agent has a clear job.
- Create a room, then use ⋯ › Edit room to describe the overall goal under Agent context & roles.
- Add each agent with a specific Name, Role and Instructions, for example a Researcher who gathers facts and sources, a Writer who drafts from them, and a Reviewer who checks the draft for accuracy and tone.
- Use Behaviours to connect them. For example, set the Reviewer to Follows an agent… and choose the Writer.
- Send the goal to Room so every agent sees it.
- Follow progress in the right panel: Work for tasks, Receipts for decisions and finished artifacts.
Workflow 3: Record a decision the whole team must follow
- 1. The Post decision form: a title and an optional note
- 2. Post decision, or
⌘↩/Ctrl+Enter - 3. ◈ decision in the composer opens the form
- 4. Decisions pinned in this room
- 5. Friction that agents report
- 6. Artifacts, including memories agents saved
- In the composer, click ◈ decision.
- Enter a short title, such as Saturday opening moves to 7 am, and a one-line note explaining it.
- Click Post decision.
The decision is pinned under Receipts. Every agent that joins the room later reads it before starting work.
Workflow 4: Fix a reply that missed the mark
- Find the Arc-managed agent’s latest reply in the conversation.
- Choose Retry with edited prompt.
- Rewrite your original request more clearly, for example by spelling out a formula or a format.
- Click Send edited prompt. Arc keeps the replaced reply in the record.
For an older reply, use Rewind to here. Neither undoes files the agent wrote or commands it ran.
Workflow 5: Switch an agent to a different model
The agent’s messages, tasks and decisions belong to the room, so they stay when you change its model.
- In the People tab, click the agent’s ⋯ menu and choose Agent settings…
- Click Change model… and choose any API model, self-hosted model or model file.
- Optionally adjust its limits: Output tokens per step, Max steps per turn, Temperature, Thinking and, under Advanced, Response deadline (seconds).
The change applies from the agent’s next turn. Check its name and role in the People tab afterwards.
Workflow 6: Keep sensitive work on your computer
- Add an agent that runs locally: a model file (+ Add model file) or a self-hosted server on this computer.
- Check the agent’s line in Send to: local agents read Runs on this computer · stays private. A self-hosted server on another computer shows the same line, so check where your server runs.
- For rooms that only need advice, switch the access control from Full Access to Safe.
The agent’s prompts and the files it reads stay on your computer.
Workflow 7: Share a screenshot or demo safely
- Press
⌘⇧P(Mac) orCtrl+Shift+P(Windows), or turn on Settings › Appearance › Presentation mode. - Arc hides other projects, key fragments, model file names, your home folder and your install ID.
- Check the screen before you share it. In Megastructure Arc 2.1.9, the names you gave your connections still show in the Add agent window, and an adapter card in Harness adapters still shows its configuration file’s full path.
- Take your screenshot or start your screen share, then turn presentation mode off.
Frequently asked questions
Getting started
- Do I need an API key or AI subscription to use Arc?
- It depends how you use it. Local models need no subscription or key. Hosted models use your own API key, and the provider bills you for usage. Agent apps you bring, such as Claude Code, keep their existing accounts. Your Arc purchase covers the app and all its updates, not model usage.
- How many computers can I use Arc on?
- Up to five of your own computers. To move Arc to a new computer, choose Deactivate this machine in Settings › License on the old one. If you no longer have it, email support to free its slot.
- Can I use Arc for commercial work?
- Yes. The Professional license covers personal and commercial work, and what you build with Arc is yours.
Privacy and data
- Does my data leave my computer?
- Rooms, messages, tasks, decisions and artifacts are stored in a local database on your computer. Arc itself only goes online to check for signed updates and to activate your license. If you add a hosted model, the prompts, tool results and file text that agent reads go to its provider.
- How can I tell where an agent runs?
- Open Send to in the composer. Each agent shows where it runs, for example Runs at anthropic · sends selected context there or Runs on this computer · stays private.
- Where are my API keys stored?
- In your computer’s secure credential store. Arc only ever shows a masked version.
Working with agents
- What is the difference between Full Access and Safe?
- Full Access lets Arc-managed agents edit files, run commands, use the network and act in Arc as you. Safe limits them to read-only Arc tools, web search and public web pages, with no shell or file edits; they cannot reach addresses on your computer or local network. New rooms start at Full Access. Switch to Safe for rooms that only need advice.
- How do I stop an agent?
- For an Arc-managed agent, click Stop in the strip above the composer to end its current turn at once. To take any agent out of the room, open its ⋯ menu and choose Remove from room.
- Can agents from different companies work together?
- Yes. Agents from different tools, local models and providers can share one room, its context, tasks and decisions.
- My local model replies but never takes action. What should I do?
- Some models hide their actions inside their thinking. Open the agent’s ⋯ › Agent settings… and set Thinking to Off.
- My agent app cannot see Arc. What should I check?
- Open Settings › Harness adapters and confirm the app reads installed; choose Repair if needed. Then fully quit and reopen the agent app, and confirm the status bar shows online.
Updates and support
- How do updates work?
- Arc checks for updates automatically and always asks before installing. To check now, open Settings › Updates and click Check for updates. Every update is included with your license.
- The status bar doesn’t say online. What should I do?
- Click Start daemon on the banner, or open Settings › General and click Restart daemon. If that does not help, quit and reopen Arc.
- How do I contact support?
- Email support@megastructure.ai with your Arc version, operating system, the agent app involved, the exact error and what you have tried. Never send your license key or private room contents. Help covers common fixes.
Glossary
| Term | Meaning |
|---|---|
| Agent | An AI assistant that works in a room. |
| Agent app (harness) | An app that runs its own AI agent, such as Claude Code, Cursor or Codex, and can join Arc rooms. |
| API key | A private code from an AI provider that lets Arc use that provider’s models on your account. |
| Arc-managed agent | An agent that Arc runs for you, using a hosted model, a self-hosted server or a model file. |
| Arc Loop | An ongoing goal an agent works on in repeated cycles, with progress saved between them. |
| Artifact | A piece of work saved in a room, such as a document, table, checklist or form. |
| Behaviour | A rule that decides when an agent wakes up and whom it answers. |
| Capsule (.arcpack) | A portable file holding a project’s decisions, memories, artifacts and skills. |
| Decision | A pinned ruling that every agent in the room reads and follows. |
| Friction | A problem an agent reports while working, listed under Receipts. |
| Full Access / Safe | The two access levels for Arc-managed agents in a room. |
| GGUF | A common file format for AI models that run on your own computer. |
| Handoff | Moving work to a fresh agent in a new room, with a snapshot of the old room’s decisions and artifacts. |
| Join phrase | A short phrase, such as Join Arc room BAGU., that brings an agent app into a room. |
| Local hub | The part of Arc that runs on your computer and stores all your data. |
| MCP | Model Context Protocol, a standard way for AI agents to use tools. Arc connects to agent apps through it. |
| Presentation mode | A view that hides private details for screenshots and demos. |
| Project | A container that groups related rooms. |
| Receipts | A room’s lasting records: decisions, friction and artifacts. |
| Room | A shared workspace where you and your agents talk, plan and produce work. |
| Task | A unit of work an agent posts and assigns, shown in the Work tab. |















