# Account, sign-in, and settings (/docs/help/account)
## Sign in with the correct account [#sign-in-with-the-correct-account]
Open [OpenMeld](https://app.openmeld.ai) and use the sign-in method associated with your account. Available methods appear on the page.
For an email link or verification code, check the inbox for the entered address. Use the current message and follow any expiry guidance. If needed, return to sign-in and request a new one.
If you use a passkey, follow your browser or device's sign-in prompt.
## Your workspace is missing [#your-workspace-is-missing]
Check the signed-in email or provider account, then open the workspace switcher. A second sign-in method can lead to a different account if it was not linked.
For a team invitation, use the invited account and check that the invitation is still valid. Ask an owner or admin to confirm membership.
## Profile and account security [#profile-and-account-security]
Open **Settings → Personal → Account** to manage the available profile, sign-in, session, and CLI-access controls.
Check active sessions and revoke an unfamiliar session through the product's controls. Keep one-time CLI tokens private.
Some account management actions open in the web app when you use Desktop.
## Appearance and privacy [#appearance-and-privacy]
Use **Appearance** for the display choices available in your client. Use **Notifications** to control attention and delivery.
Review **Privacy** and Center Agent memory controls for the context choices available to your account. Appearance settings do not change sharing permissions.
## Delete an account [#delete-an-account]
Account deletion is different from signing out or leaving one Organization. In Account settings, **Delete account…** opens a review of what will be deleted and what remains with your teams.
Read that review before continuing. Ownership or transfer requirements can block deletion. Shared work in a surviving team is not necessarily erased when your account closes.
Do not use account deletion to troubleshoot sign-in, downloads, or a disconnected computer.
Next: [Workspaces and members](/docs/team/workspaces).
# An Agent is not replying (/docs/help/agent-not-replying)
## 1. Confirm that you asked it to work [#1-confirm-that-you-asked-it-to-work]
Open the original message. Check the intended Agent and its addressing mode.
**Mention only** does not start work. Adding an Agent to a Space or assigning a Task also does not prove an execution started.
If needed, send one clear request to the intended Agent. Avoid repeatedly sending the same task.
## 2. Read the current status [#2-read-the-current-status]
* **Checking** means availability is not yet known.
* **Offline** can indicate that the required local connection is unavailable.
* **Unavailable** needs the specific reason shown on that Agent.
* An active work indicator means you should inspect that work, not create a duplicate request.
Use the message on screen rather than treating all failures as the same connection problem.
## 3. For a Local Agent [#3-for-a-local-agent]
Check the computer assigned to the Agent, not only the computer in front of you.
1. Make sure it is awake and connected.
2. Open its Local agents details.
3. Resolve sign-in, disabled-tool, update, or installation problems.
4. Check the selected Project and working folder.
5. Return to the Agent and check availability again.
See [Computer help](/docs/help/computer). A working Service connection alone does not prove the chosen agent tool is ready.
## 4. For a hosted Agent [#4-for-a-hosted-agent]
Check its model, workspace access, usage, and any pause or configuration notice. Connecting your laptop does not repair hosted model or account failures.
An owner or admin may need to resume a paused Center Agent.
## 5. Check for a question or result [#5-check-for-a-question-or-result]
The Agent may be waiting for information, approval, or review. Check [Inbox](/docs/help/notifications) and the original thread.
Look for a result, an explicit failure, or a stopped outcome. Activity metadata and a recent timestamp are not completion evidence.
## Before retrying [#before-retrying]
Inspect files and external destinations for partial work. If the outcome remains unclear, record the visible error and time, then [send feedback](/docs/help/troubleshooting#ask-for-help).
# Fix a computer connection (/docs/help/computer)
## The computer is disconnected [#the-computer-is-disconnected]
Make sure the named computer is powered on, awake, and online. Open the app's **Search or run a command**, search for **Computers**, and inspect that computer.
Use its reconnect or setup action. Run the new command on that machine, under the intended operating-system user.
Do not connect a second machine and assume it repaired the first.
## The setup command expired [#the-setup-command-expired]
Return to the OpenMeld setup page and get a fresh command. Short-lived authorization is expected to expire.
Do not paste someone else's command or reuse one from a screenshot.
## The command is not found [#the-command-is-not-found]
Open a new Terminal or PowerShell window after installation. Confirm you used the command for that operating system.
If the problem continues, keep the exact error and follow the installer guidance. See [Terminal basics](/docs/start/terminal).
## The computer connects but no Agent is ready [#the-computer-connects-but-no-agent-is-ready]
Local tool discovery can continue after the computer connects. Open **Local agents** and inspect the tool you intend to use.
Install or sign in to the underlying tool if required. An explicitly disabled tool stays disabled until you choose to enable it.
Then check the Agent Profile's computer, tool, model, and folder.
## Tools need an update [#tools-need-an-update]
Use the offered update action. A background update starting is not the same as completion. Check its final status before expecting a new capability.
If the update reports a failure or restoration, preserve that message and follow the recovery action.
## Read-only checks in the terminal [#read-only-checks-in-the-terminal]
If the CLI is installed, these commands can help identify the affected layer:
```sh
openmeld service status
openmeld doctor --view agent
```
The default doctor path reports diagnostics. `--repair` is a different, mutating action; do not add it merely to collect a report.
Do not publish raw diagnostic output without checking it for private information.
## A folder is unavailable [#a-folder-is-unavailable]
Open the [Project connection](/docs/work/projects). Confirm the folder still exists on the assigned computer and update the connection if it moved.
# Desktop installation and updates (/docs/help/desktop)
## Choose the matching build [#choose-the-matching-build]
Use the [official download page](https://app.openmeld.ai/desktop). The current preview lists Apple silicon, Windows x64, and Windows ARM64 selectors.
Download only a build that the page actually offers. If your platform has no published build, use the browser.
On a Mac, check **Apple menu → About This Mac**. On Windows, check **Settings → System → About**.
## The download or installer failed [#the-download-or-installer-failed]
Retry from the official page. Check available disk space and the exact installer error.
On a Mac, open the DMG and drag the app into Applications before launching it. On Windows, open the installer and follow its instructions for your account.
If the operating system reports a security or signature problem, keep the warning and report it. Do not disable security protections or use an unofficial repackaged download.
## Sign-in did not finish [#sign-in-did-not-finish]
Use the same account as OpenMeld Web. Complete the browser sign-in step and return to Desktop when prompted.
If the desktop sign-in flow cannot finish, check whether the same account can use [the web app](https://app.openmeld.ai). This distinguishes account access from a Desktop handoff problem.
## Updates [#updates]
Review the available update controls in Desktop. Installation, download, restart, and successful launch are separate steps.
The current Windows x64 internal preview does not support automatic updates. Check the official download page for a newer build. Use the browser if your organization does not permit unsigned preview software.
After updating, open the app and check that your workspace loads. Do not assume a downloaded update is already running.
Desktop and OpenMeld Tools can have separate update states. Inspect the connected computer when local agent work is the problem.
## Keep working in the browser [#keep-working-in-the-browser]
You can use the web app while a Desktop issue is investigated. Local execution still depends on the assigned computer and its Tools connection.
When reporting a problem, include the operating system, processor type, Desktop version, error, and time. See [Find a fix](/docs/help/troubleshooting).
# Inbox, search, and notifications (/docs/help/notifications)
Open **Inbox** from the app navigation or bottom toolbar.
## Choose what to review [#choose-what-to-review]
* **All** includes your available updates.
* **Needs action** focuses on work waiting for your input.
* **Needs review** focuses on decisions and approvals.
* **Unread only** narrows the current view to unread items.
Open an item to return to the original conversation or work. Check that source before deciding what to do.
## Read is not resolved [#read-is-not-resolved]
Opening an item marks it read. It does not approve a plan, answer a question, resolve a failure, or complete a Task.
**Mark all as read** can include items outside the current filter. Read its confirmation. Pending actions and reviews remain open.
Dismissing an Inbox item removes it from your attention list; it does not necessarily resolve or delete the underlying work.
## Search [#search]
Use Inbox search for your items. Use **Search or run a command** for navigation and available workspace message search.
Search results preserve access rules. If a source is no longer available, ask its owner to check access.
An incomplete or failed search is not proof that no matching message exists.
## Control notifications [#control-notifications]
Open **Settings → Personal → Notifications**. Review the available delivery settings. Use a conversation's notification controls for its local preference.
Browser notifications require browser permission. Desktop notifications also depend on operating-system settings. Allowing notifications does not give a website access to your local files.
If notifications do not appear, check app preferences, browser or OS permission, and whether the conversation is muted.
## Recommendations [#recommendations]
Center Agent can place a suggestion in Inbox so you can return to it. Accepting a recommendation and authorizing its proposed work are separate decisions.
Next: [Messages and threads](/docs/work/messages).
# Find a fix (/docs/help/troubleshooting)
| What happened | Start here |
| --------------------------------------------------- | --------------------------------------------------- |
| You cannot sign in or find your workspace | [Account and sign-in](/docs/help/account) |
| Setup does not connect your computer | [Computer connection](/docs/help/computer) |
| An Agent is offline, unavailable, or does not reply | [Agent not replying](/docs/help/agent-not-replying) |
| Desktop will not install or open | [Desktop help](/docs/help/desktop) |
| You cannot find a notification or approval | [Inbox and notifications](/docs/help/notifications) |
| A tool asks for access or shows a connection error | [Plugins](/docs/team/plugins) |
| A scheduled task did not produce a result | [Routine history](/docs/work/routines) |
| A file or conversation is unavailable | [Access and sharing](/docs/team/access) |
## Check these first [#check-these-first]
1. Confirm the signed-in account and workspace.
2. Read the exact message on the affected item.
3. Check the task's original conversation or result destination.
4. Follow the recovery action attached to that failure.
A connected computer, a sent message, and a completed task are different facts. Check the one that matches your problem.
## Avoid duplicate actions [#avoid-duplicate-actions]
Before retrying a task that sends messages, creates issues, or changes files, inspect its destination. The external action may have succeeded even if a later progress update failed.
Do not reset your account, clear local application data, or reinstall everything as a first response. Those actions can remove useful context without fixing the cause.
## Ask for help [#ask-for-help]
If **Send feedback** is available, open **Settings → Personal → Send feedback**. Describe:
* What you tried to do.
* What you expected.
* What actually happened.
* The time and timezone.
* The relevant app version, platform, and visible error.
Include a permitted link to the affected item when useful. Remove setup grants, passwords, API keys, private messages, and unrelated personal information.
Sending feedback records the report. It does not guarantee an immediate reply or start an Agent automatically.
# OpenMeld guides (/docs)
OpenMeld brings people and AI teammates into shared work. You can start in your browser, ask Center Agent for help, and keep the conversation and result together.
## New here? [#new-here]
Start with one small task. You do not need to install a local agent to try a hosted conversation.
Sign in, send your first request, and check the answer.
Choose the browser, Desktop preview, or computer setup.
Accept an invitation and find the right workspace.
Use the local agents and files on your own machine.
## Make something useful [#make-something-useful]
* [Work with Center Agent](/docs/work/center-agent): turn context into a clear result.
* [Create an Agent](/docs/work/agents): choose a local or hosted teammate.
* [Work in a Space](/docs/work/spaces): bring people and agents into one conversation.
* [Schedule a Routine](/docs/work/routines): run a well-defined task later or repeatedly.
## Bring the right context [#bring-the-right-context]
* [Files](/docs/work/files) for the source material and artifacts.
* [Plugins](/docs/team/plugins) for authorized external tools.
* [Skills](/docs/team/skills) for repeatable instructions.
* [Team Knowledge](/docs/team/knowledge) for approved reference material.
## Find an answer [#find-an-answer]
Use search at the top of the sidebar. Search for what you want to do, such as “connect computer,” “invite,” or “agent not replying.”
Browse [all features](/docs/reference/feature-guide), look up [product terms](/docs/start/concepts), or start with [troubleshooting](/docs/help/troubleshooting).
The guides describe user-facing workflows. Available actions depend on your role, connected tools, and installed version. Desktop downloads remain labeled Preview until their release status changes.
# Use OpenMeld from a terminal (/docs/reference/cli)
The OpenMeld CLI is the command-line interface. It lets people and local agents work with OpenMeld from a terminal.
If you are new to terminals, start with [Connect your computer](/docs/start/connect-computer). Use the setup command generated for your account and operating system. Do not install a similarly named package from an unrelated source.
## Check the installation [#check-the-installation]
Open a new terminal after setup:
```sh
openmeld --version
openmeld --help
```
The first command prints the installed version. The second shows its supported commands.
## Common entry points [#common-entry-points]
| Command | Purpose |
| ------------------------------ | ------------------------------------------------- |
| `openmeld start` | Enter the guided starting workflow. |
| `openmeld setup` | Set up or align this computer with your account. |
| `openmeld service status` | Inspect the local Service state. |
| `openmeld doctor --view agent` | Read structured setup and diagnostic information. |
Setup can change local configuration. Doctor's default diagnostic path is read-only; repair options are separate.
## Discover the exact command [#discover-the-exact-command]
Use the help provided by your installed version:
```sh
openmeld files --help
openmeld skills --help
openmeld routines --help
```
Files, Skills, and Routines remain subject to the same workspace and permission rules as the app. A command existing does not grant access.
## Human and Agent views [#human-and-agent-views]
Human output is intended for a person reading a terminal. Supported commands can expose `--view agent` for structured output and explicit automation arguments.
Use each command's help for required flags. Do not turn an example identifier into a real Organization, Space, or file target without checking it.
## Keep secrets out of shared material [#keep-secrets-out-of-shared-material]
Setup commands can include short-lived authorization. CLI diagnostics can include operational context. Review output before pasting it into a shared conversation or support report.
For local problems, see [Computer help](/docs/help/computer).
# Read these docs with an Agent (/docs/reference/docs-for-agents)
These docs are public. They contain product guidance, not access to your workspace, conversations, files, or accounts.
## Choose a format [#choose-a-format]
* [Documentation index](/llms.txt): find the relevant guide.
* [Complete documentation](/llms-full.txt): read the public guide collection.
* Use **Copy Markdown** on a guide to copy its text.
* Use the page's **Open** menu for its Markdown view.
HTML and Markdown come from the same MDX source.
## Read one guide [#read-one-guide]
Append `.md` to a guide path. For example:
```sh
curl -fsSL https://docs.openmeld.ai/docs/start/quickstart.md
```
The guide overview is at [/docs/index.md](/docs/index.md). Requests for a guide with `Accept: text/markdown` redirect to its Markdown representation.
## Use docs as context, not permission [#use-docs-as-context-not-permission]
A guide explains how a feature works. It does not authorize an Agent to create accounts, connect tools, send messages, change files, or disclose private data.
Check the user's request, current account, workspace, and product permissions before acting. Read the actual UI or command help for version-sensitive details.
## Check your result [#check-your-result]
If a guide directs you to create an artifact, verify that artifact at its intended destination. A successful command exit, saved draft, or progress message may prove only part of the task.
For troubleshooting, preserve the visible failure and avoid duplicating a consequential action whose outcome is unknown.
# Find a feature (/docs/reference/feature-guide)
Choose the outcome you need. Each guide explains the relevant setup, workflow, and limits.
## Start and install [#start-and-install]
| Feature | Guide |
| ---------------------------------------------------- | ----------------------------------------------------- |
| Sign-up, first workspace, first Center Agent request | [Get started](/docs/start/quickstart) |
| Browser, Mac and Windows Desktop preview | [Downloads](/docs/start/downloads) |
| Computer setup and OpenMeld Tools | [Connect your computer](/docs/start/connect-computer) |
| Terminal and PowerShell basics | [Open a terminal](/docs/start/terminal) |
| Invitations and joining an Organization | [Join your team](/docs/start/join-team) |
| Product vocabulary | [A few useful terms](/docs/start/concepts) |
## Conversations and agent work [#conversations-and-agent-work]
| Feature | Guide |
| ----------------------------------------------------------- | --------------------------------------------- |
| Center Agent, separate conversations, progress, outcomes | [Center Agent](/docs/work/center-agent) |
| Main navigation, quick controls, Graph and Desk | [Find your way around](/docs/work/navigation) |
| Hosted file and command work, persistence and output | [Cloud Computer](/docs/work/cloud-computer) |
| Local and Remote Agents, models, execution settings | [Agents](/docs/work/agents) |
| Shared Spaces, membership, guide, pins, Graph | [Spaces](/docs/work/spaces) |
| Direct messages, threads, Wake, Mention only, reactions | [Messages](/docs/work/messages) |
| Deliverables, assignees, reviewers, status, issue links | [Space Tasks](/docs/work/tasks) |
| Team, private and Space files, upload, copy, history, Trash | [Files](/docs/work/files) |
| Projects, local folders, defaults and worktrees | [Projects](/docs/work/projects) |
| Schedules, Run history, result destinations and pause | [Routines](/docs/work/routines) |
| Local session awareness and sharing limits | [Agent Activity](/docs/work/activity) |
| Member memory and Organization memory | [Memory](/docs/work/memory) |
## Team and tools [#team-and-tools]
| Feature | Guide |
| ------------------------------------------------------------------- | ----------------------------------- |
| Personal Workspace, Organization, invitations and roles | [Workspaces](/docs/team/workspaces) |
| Slack, Lark, Feishu and account linking | [Channels](/docs/team/channels) |
| Personal and shared app accounts, GitHub, connection requests | [Plugins](/docs/team/plugins) |
| Create, install, maintain, enable and remove Skills | [Skills](/docs/team/skills) |
| Team Knowledge, review, approved guidance and Profile Reports | [Knowledge](/docs/team/knowledge) |
| Sharing, tool identity, permission modes and information protection | [Access](/docs/team/access) |
| Hosted and local provider usage | [Usage](/docs/team/usage) |
## Settings, help, and reference [#settings-help-and-reference]
| Feature | Guide |
| ----------------------------------------------------------------- | --------------------------------------------------- |
| Sign-in, profile, sessions, appearance, privacy, account deletion | [Account and settings](/docs/help/account) |
| Inbox, action and review items, search, notification delivery | [Inbox and notifications](/docs/help/notifications) |
| Local and hosted Agent recovery | [Agent not replying](/docs/help/agent-not-replying) |
| Computer connection, tool discovery, diagnostics and updates | [Computer help](/docs/help/computer) |
| Desktop install, sign-in and update recovery | [Desktop help](/docs/help/desktop) |
| Error reporting and feedback | [Find a fix](/docs/help/troubleshooting) |
| Terminal commands and Agent View | [CLI](/docs/reference/cli) |
| Markdown, documentation index and machine-readable guides | [Docs for agents](/docs/reference/docs-for-agents) |
## Availability [#availability]
The app is the source for your current role, connections, models, and available actions. A disabled control or a Request action is not a connected capability.
Some navigation changes arrive before every installed client updates. Routines may appear in Center Agent or Team. Desktop is a Preview, and each platform has its own download availability.
These docs do not present internal experiments, unshipped roadmap items, or a private service API as supported public features. For a release's actual changes, read the [Changelog](https://openmeld.ai/changelog).
# A few useful terms (/docs/start/concepts)
You do not need to memorize these terms. Use this page when a label is unfamiliar.
| Term | What it means | Example |
| ------------- | ------------------------------------------------------------ | ------------------------------------------------------- |
| Workspace | The personal or shared area you are currently using. | Your Personal Workspace or your company's Organization. |
| Organization | A shared workspace with members and managed access. | A team working on a product launch. |
| Center Agent | The workspace's hosted AI teammate. | Ask it to organize notes or help coordinate work. |
| Agent Profile | A named AI teammate you can talk to and add to a Space. | A Local Agent that uses your Codex installation. |
| Local Agent | An agent that runs through a connected computer. | Review files in a local project folder. |
| Remote Agent | A hosted agent that does not need your local computer. | A separate AI teammate with its own model setting. |
| Space | A shared conversation with people, agents, and related work. | A launch planning room. |
| Wake | Asking an agent to do work through a message. | Address an Agent and ask it to review a draft. |
| Mention only | Refer to an agent without asking it to run. | Tell a teammate which Agent you plan to use later. |
| Project | A shared project identity that can connect to local folders. | The same repository on two people's computers. |
| Plugin | A connection or capability agents use while working. | An authorized GitHub connection. |
| Channel | A messaging service where you talk to Center Agent. | Slack, Lark, or Feishu. |
| Skill | Reusable instructions and resources for a kind of work. | Your team's release-review checklist. |
| Routine | Instructions scheduled to run once or repeatedly. | Prepare a weekly update. |
| Run | One recorded occurrence of a Routine. | Monday's weekly update. |
## Three distinctions that prevent mistakes [#three-distinctions-that-prevent-mistakes]
**An Agent is not a computer.** Several Agent Profiles can use a computer, and hosted agents do not require yours.
**A reply is not always completed work.** Read the result and check the requested file, change, or destination.
**A shared location changes who can see information.** Check the audience before you attach a file, copy content, or select a Routine's result destination.
Next: [Get started](/docs/start/quickstart).
# Connect your computer (/docs/start/connect-computer)
Connect a computer when you want a local agent to work there. You can use hosted Center Agent conversations without this step.
## Before you start [#before-you-start]
Use the computer where the work should run. Sign in to the correct OpenMeld account. If you want to use Codex, Claude Code, Cursor, Grok Build, or OpenCode, install and sign in to that tool as well. OpenMeld does not replace the tool's account or subscription.
## 1. Open computer setup [#1-open-computer-setup]
In Center Agent, choose **Connect my computer**. You can also use the computer setup step during onboarding. To manage existing connections, open **Search or run a command** and search for **Computers**.
If OpenMeld shows an existing computer, check its name. A connected computer in your account might be a different machine.
## 2. Copy your setup command [#2-copy-your-setup-command]
Choose the correct operating system and copy the command shown in OpenMeld. This command connects the computer to your signed-in account.
Keep this page open. Do not share the command: it can contain a short-lived account authorization. If it expires, get a new command from the setup page.
## 3. Run the command [#3-run-the-command]
Open [Terminal or PowerShell](/docs/start/terminal). Paste the complete command and press Enter.
Follow the instructions in the terminal. Return to the setup page when it reports that the computer is connected.
## 4. Check the result [#4-check-the-result]
Look for **Computer connected** and confirm the computer name. Local agents can continue syncing in the background after this appears.
Open the computer's details to check its **Local agents**. A tool that needs sign-in or repair stays unavailable until you resolve that condition. Finding the tool on disk is not enough.
## 5. Create or select an Agent [#5-create-or-select-an-agent]
An Agent Profile is the named teammate you talk to in OpenMeld. [Create a Local Agent](/docs/work/agents), choose this computer and its local agent tool, then select the intended folder or Project.
Before assigning work, check the Agent's current availability.
## Keep the computer available [#keep-the-computer-available]
Local work needs the computer and its connection. Sleep, shutdown, network loss, or signing out of the underlying agent tool can interrupt availability. Hosted agents have a different execution path.
Connected computers can keep OpenMeld Tools up to date. Review **Keep OpenMeld Tools up to date** in the computer's Tools card.
If setup fails, use [Computer connection help](/docs/help/computer). Avoid resetting your local state as a first step.
# Downloads and installation (/docs/start/downloads)
## Choose what you need [#choose-what-you-need]
| You want to… | Start here |
| ------------------------------------------------- | ------------------------------------------------------------------- |
| Talk to Center Agent or collaborate in Spaces | [Open the web app](https://app.openmeld.ai). No download is needed. |
| Use OpenMeld in a desktop window | [Download Desktop preview](https://app.openmeld.ai/desktop). |
| Let local agents work with files on your computer | [Connect your computer](/docs/start/connect-computer). |
| Use OpenMeld from a terminal | [Use the CLI](/docs/reference/cli). |
**Desktop and OpenMeld Tools do different jobs.** Desktop gives you an app window. Tools connect your computer and carry local agent work. Signing in to Desktop alone is not proof that a local agent can run.
## Install Desktop on a Mac [#install-desktop-on-a-mac]
The published Mac preview targets **Apple silicon**: Macs with an M-series chip.
1. Open Apple menu → **About This Mac**. Check the **Chip** field.
2. Open the [Desktop download page](https://app.openmeld.ai/desktop).
3. Choose **Mac (Apple silicon)** and download the DMG.
4. Open the DMG. Drag **OpenMeld Desktop** to **Applications**.
5. Open it from Applications and sign in with your OpenMeld account.
Use the browser on an Intel Mac unless the download page offers a compatible build.
## Install Desktop on Windows [#install-desktop-on-windows]
1. Open Windows **Settings → System → About** and check **System type**.
2. Open the [Desktop download page](https://app.openmeld.ai/desktop).
3. Choose **Windows (x64)** or **Windows (ARM64)** to match your computer.
4. If a download is available, open the EXE and install it for your Windows account.
5. Open OpenMeld from the Start menu and sign in.
The Windows x64 download is currently an unsigned internal preview, with no automatic updates. The ARM64 selector currently shows no download. Check the page for the latest availability and release notes. If your build is unavailable, use [OpenMeld Web](https://app.openmeld.ai).
## Linux and other devices [#linux-and-other-devices]
Use the web app. The current Desktop download page does not offer a Linux installer or native mobile app. To connect a supported computer for local work, follow the platform-specific command in [computer setup](/docs/start/connect-computer).
## Check the installation [#check-the-installation]
You should see your signed-in workspace. Confirm the account and workspace name before you start work.
Desktop is currently a **Preview**. Use the download page for the available version and platform status. If installation or sign-in fails, see [Desktop help](/docs/help/desktop).
# Join your team (/docs/start/join-team)
You need an invitation or a valid join link for a private Organization. An Organization is the shared workspace that owns your team's members, agents, tools, and files.
## Accept an invitation [#accept-an-invitation]
1. Open the invitation email or join link your teammate sent.
2. Sign in to OpenMeld. For an email invitation, use the invited account.
3. Check the Organization name before accepting.
4. Complete any required account setup.
5. Use the workspace switcher to confirm you are in the right Organization.
If the invitation expired or was withdrawn, ask an owner or admin for a new one. Creating another account does not repair an invitation to your existing account.
## Find your first conversation [#find-your-first-conversation]
Open **Spaces** to see conversations available to you. If someone invited you to a specific Space, open that invitation or ask them for its link.
Organization membership and Space access are separate. Joining a team does not make every private conversation visible.
Open **Team → People** to find teammates. You can start a direct message from a person's profile or the conversation controls.
## If your team works in Slack, Lark, or Feishu [#if-your-team-works-in-slack-lark-or-feishu]
Your administrator may already have connected Center Agent. You can ask it for help in that connected workspace. Some actions need you to connect your OpenMeld account.
Use the private **Connect my account and continue** action when it appears. Check both accounts before confirming. See [Channels](/docs/team/channels).
## You do not need to connect a computer to join [#you-do-not-need-to-connect-a-computer-to-join]
Connect a computer only when you want to run your local agents. Joining a Space and making a local agent available are different steps.
Next: [Work in a Space](/docs/work/spaces).
# Get started (/docs/start/quickstart)
OpenMeld is a place to work with people and AI teammates. Start with **Center Agent**, the assistant in your workspace. Give it a small task, check the answer, and then connect more tools when you need them.
## 1. Open your workspace [#1-open-your-workspace]
[Open OpenMeld](https://app.openmeld.ai). Sign in, or choose **Sign up** to create an account. Follow the sign-in and email checks shown on the page.
During setup, choose:
* **Personal Workspace** to try OpenMeld on your own.
* An existing **Organization** if your team invited you.
* **Create Organization** to set up a shared workspace.
You can skip optional connections during setup. Skipping a connection does not connect that tool. You can add it later.
## 2. Open Center Agent [#2-open-center-agent]
Complete setup and open Center Agent. If you are already in the app, use **Open Center Agent** in the bottom toolbar. Open the full chat when you want more room.
The first conversation offers **Show me around**. Use it for a guided introduction, or send a task of your own.
## 3. Try a small task [#3-try-a-small-task]
Paste this example into the message box, then send it:
> Turn these notes into a short meeting agenda: we need to choose a launch date, assign someone to test the signup flow, and decide who will write the announcement. Keep it to three agenda items. Do not send or publish anything.
This example needs no connected account, local files, or installed coding agent.
## 4. Check the answer [#4-check-the-answer]
Read the reply in the same conversation. Check that it includes all three topics and does not invent names or dates.
Continue with:
> Add one question we should answer for each agenda item.
You now have an agenda you can review and use. For your next task, add only the tools or files that task needs.
## Choose your next step [#choose-your-next-step]
Keep a shared task and its result in one Space.
Connect the computer that has your files and agent tools.
Let an authorized agent work with the tools you already use.
If the reply fails, follow the message on screen and use [Agent not replying](/docs/help/agent-not-replying). Do not repeat a request that could send a message or change data until you check its outcome.
# Open Terminal or PowerShell (/docs/start/terminal)
A terminal is an app where you type commands. For OpenMeld setup, you only need to paste the command from your setup page and press Enter.
## Mac [#mac]
1. Press **Command + Space** to open Spotlight.
2. Type **Terminal** and press Enter.
3. Copy the setup command from OpenMeld.
4. Click the Terminal window, press **Command + V**, then press Enter.
Leave the window open while setup runs. Read any prompt before answering it.
## Windows [#windows]
1. Open the Start menu.
2. Search for **PowerShell** and open it.
3. Copy the **Windows** setup command from OpenMeld.
4. Paste it into PowerShell and press Enter.
Use the shell named by the setup instructions. A Mac command pasted into PowerShell will not work. Do not open an administrator terminal unless the installer specifically requires it.
## Linux [#linux]
Open your distribution's Terminal app. Copy the matching command from OpenMeld, paste it, and press Enter.
## What the text means [#what-the-text-means]
* **A command** tells the computer what to do.
* **Output** is the text the command prints. Do not paste that output back as another command.
* **A prompt** means the terminal is waiting for your next input.
* **Command not found** means the shell cannot find the named program.
When setup finishes, return to OpenMeld and check that the intended computer is connected. If the command is not found after installation, open a new terminal window before trying again.
## Copy safely [#copy-safely]
Copy the command itself, not Markdown backticks or a leading prompt symbol such as `$`. Use the complete line. Keep setup authorization codes private.
Next: [Connect your computer](/docs/start/connect-computer).
# Access, sharing, and permissions (/docs/team/access)
OpenMeld separates workspace membership, conversation access, file location, and tool permissions. Check each one for the task you are doing.
## Check the audience first [#check-the-audience-first]
Before sending a message or attaching a file, check the active workspace and conversation members.
A team member may be allowed to read Team files but not a private Space. A Space guest may see a copied Space file without gaining access to all Team files.
Copying content into a broader audience is a sharing action. Later deleting the source does not remove independent copies.
## Distinguish discussion from execution [#distinguish-discussion-from-execution]
A human mention draws attention. **Mention only** refers to an Agent. A **Wake** asks an Agent to work.
Inviting an Agent, assigning a Task, or accepting a recommendation does not grant unlimited permission to make changes.
State the action, target, and limits when asking for consequential work.
## Review tool identity [#review-tool-identity]
For external actions, check the connected account or Organization identity. Personal and shared credentials have different owners.
An admin role does not automatically grant access to another person's private agent context or personal account.
## Local Agent permission settings [#local-agent-permission-settings]
A Local Agent uses its selected tool, computer, working folder, and permission mode. Review those choices before allowing it to change files or run commands.
A connected computer is not proof that every tool is signed in or that a specific Agent is available.
## Information protection [#information-protection]
Owners and admins can add classification hints under **Settings → General → Information protection**. Use category names or document types, not secret values or sensitive records.
These hints guide information handling. They do not grant tool access or replace the built-in protection rules.
For legal terms and data policy, read the current [Privacy Policy](https://openmeld.ai/privacy) and [Terms](https://openmeld.ai/terms). This guide explains product controls; it does not add contractual guarantees.
## When access changes [#when-access-changes]
A removed member or disconnected tool can make an old link unavailable. Ask the owning person or admin to review access. Do not share session cookies, setup grants, or credentials as a workaround.
# Connect Slack, Lark, or Feishu (/docs/team/channels)
A Channel is a messaging entrance to Center Agent. It is different from a [Plugin](/docs/team/plugins), which lets agents use an external tool while doing work.
## Before connecting [#before-connecting]
You need a shared Organization. Personal Workspaces show Channel options for discovery but cannot connect a workspace-level Channel.
Use an account with the required OpenMeld and provider administration permissions. Check that you are connecting the intended messaging workspace.
## Connect Slack [#connect-slack]
1. Open **Center Agent → Channels**.
2. Choose Slack's connection action.
3. Review the workspace and permissions in Slack.
4. Complete the provider flow.
5. Return to OpenMeld and confirm the connected status.
Closing the authorization window is not connection success. If setup fails, use the status and recovery action shown in Channels.
## Connect Lark or Feishu [#connect-lark-or-feishu]
Choose the matching provider in **Channels** and follow its setup instructions. Complete all requested provider configuration.
A connection can remain **configuring** until its event and message-card callbacks are verified. Do not treat an app created in the provider console as proof that messages reach OpenMeld.
## Try the connection [#try-the-connection]
Open the connected messaging service and send Center Agent a small, non-sensitive request. Check that the reply returns to the same conversation.
For a first test, ask it to turn three notes into a checklist. Test tool actions separately after the basic message path works.
## Connect your identity when needed [#connect-your-identity-when-needed]
An internal member can use Organization-shared capabilities without first linking an OpenMeld account. Personal resources and role-sensitive actions can require a verified account connection.
When prompted, open **Connect my account and continue**. Check the messaging account and signed-in OpenMeld account before confirming. The continuation should return to the original conversation.
## Keep each connection separate [#keep-each-connection-separate]
One Organization can connect different supported providers. Repairing or disconnecting one does not disconnect the others.
Connection availability and provider permissions can change. Use the live Channels page for the current state.
# Share approved team knowledge (/docs/team/knowledge)
Use shared knowledge for information that should guide work repeatedly: a team brief, an approved process, or a maintained reference.
**Files and knowledge are not interchangeable.** Uploading a Markdown file does not automatically make it approved knowledge.
## Find Team Knowledge [#find-team-knowledge]
Open **Team → Files** and look for **Team Knowledge**. Managed knowledge and reports keep their own editing and review rules.
All current Organization members can read approved shared knowledge. Management actions depend on role and content type.
## Write useful reference material [#write-useful-reference-material]
Keep one subject per document. State the owner, purpose, and current rules. Prefer concrete instructions over broad claims.
For example:
> Customer updates must distinguish shipped features from planned work. Link the release note for each shipped feature. Do not include customer names without permission.
Do not place credentials or private personal records in a shared reference.
## Review before approval [#review-before-approval]
Owners and admins can review and manage shared knowledge through its available actions. A draft is not an approved instruction.
Check the exact saved content before approving. When a document changes, review the updated version rather than assuming an earlier approval still applies.
## Use knowledge in a task [#use-knowledge-in-a-task]
Name the relevant reference and ask the Agent to read it before applying it. If the Agent cannot access it, provide a permitted source or resolve access first.
Current approved guidance should take precedence over an old remembered rule.
## Profile Reports [#profile-reports]
Managed Profile Reports can have special ownership and edit rules. A member's ability to edit the body of their own report does not grant permission to rename, move, approve, or edit everyone else's reports.
Managed folders may not allow ordinary file actions. Use their owning workflow instead of trying to bypass the restriction.
Next: [Memory](/docs/work/memory) and [Skills](/docs/team/skills).
# Connect Plugins and accounts (/docs/team/plugins)
Plugins let authorized agents read or act in external systems. Connecting a Plugin does not authorize every possible action. Your request and the tool's permissions still matter.
## Choose personal or shared access [#choose-personal-or-shared-access]
**Your connection** uses your external account for authorized work on your behalf.
**Organization access** is a shared connection managed by an owner or admin. Review who connected it and what resources it can reach.
For GitHub, Organization access uses the OpenMeld GitHub App and acts as the OpenMeld bot. It is distinct from **Your GitHub**. Name the intended identity when that matters.
## Connect a supported tool [#connect-a-supported-tool]
1. Open **Team → Plugins** or the relevant connected-app setting.
2. Search for the tool.
3. Read the available connection options and their scope.
4. Start the connection and review provider permissions.
5. Return to OpenMeld and check that it is connected.
The catalog may include tools that offer **Request** instead of **Connect**. Requesting a tool does not make it available.
## Test with a read [#test-with-a-read]
Start with a bounded task:
> Using the Organization GitHub connection, list the open issues in the repository I name. Do not change anything.
Check that the result names the correct account or Organization and resource. Then request a specific write if you need one.
Do not paste provider credentials into a conversation to repair a connection.
## Work through Channels and Routines [#work-through-channels-and-routines]
Personal connections require the correct verified member identity. A Slack account with a matching email is not automatically linked.
GitHub work in Routines uses Organization GitHub App access. Do not assume a Routine can borrow your personal GitHub connection. Some destructive operations are unavailable for unattended work.
## Disconnect or repair [#disconnect-or-repair]
Use the connection's management action. You may need its owner or an Organization admin.
If a refresh fails, retry the connection check before making another change. A previously visible connection is not proof of current authorization.
# Use and manage Skills (/docs/team/skills)
A Skill packages a procedure with the reference material it needs. For example, a release-review Skill can explain the checks your team expects and include a report template.
A Skill is not a credential or a new permission grant.
## Find and use a Skill [#find-and-use-a-skill]
1. Open **Team → Skills**.
2. Search for the relevant Skill.
3. Read its purpose, instructions, and resources.
4. Check whether it is enabled.
5. Ask Center Agent to use it for a concrete task.
Name the input and desired result. “Use our release-review Skill on this draft” is more useful than “Use some Skills.”
## Create a Skill [#create-a-skill]
Owners and admins can create Skills.
Choose the create action, then provide a name, description, and instructions. State when the Skill applies, what inputs it needs, what steps to follow, and how to check the result.
Choose **Create skill** to save it. Opening or filling the form alone does not create a Skill.
## Install from GitHub [#install-from-github]
Use the install action and provide the source URL. If the repository contains several Skills, select one.
Review the repository, path, pinned commit, resources, and warnings before choosing **Install skill**. Treat third-party instructions and scripts as material to review, not as trusted commands merely because they are packaged as a Skill.
## Edit and maintain [#edit-and-maintain]
Owners and admins can manage any Skill. An assigned **Skill Maintainer** can edit and save that Skill, but cannot thereby install, remove, or change enablement for other Skills.
Wait for **Saved**. If saving fails, keep your edits and resolve the error before leaving.
Saved releases preserve the instructions and resources used at that revision. Check history when behavior changes.
## Disable or remove [#disable-or-remove]
Disabling stops active use while retaining the managed Skill. Removal requires confirmation and should be deliberate.
A Skill cannot make an unavailable Plugin, computer, or model available. Check those dependencies separately.
# Check usage and availability (/docs/team/usage)
OpenMeld can use hosted models and agents installed on connected computers. These can have different usage limits and accounts.
## Check the relevant usage view [#check-the-relevant-usage-view]
Open your workspace's **Settings → Usage** for the usage information available to your role.
For a Local Agent, inspect its computer and tool details. The underlying provider may also show usage in its own app or account page.
Check the period, account, and scope shown beside a number. A remaining amount for one provider does not describe every Agent.
## Hosted versus local usage [#hosted-versus-local-usage]
A hosted Agent depends on its configured model and OpenMeld availability.
A Local Agent can depend on a separate subscription, API account, or provider quota. Installing OpenMeld does not transfer that subscription or remove its limits.
Usage that is unavailable or still checking is unknown. Do not interpret it as zero usage or unlimited capacity.
## When a limit stops work [#when-a-limit-stops-work]
1. Read the actual failure message.
2. Identify the affected Agent, model, or account.
3. Check the matching usage or provider setting.
4. Wait for the stated reset, repair the account, or select another permitted configuration.
5. Retry only after checking whether the original task already produced a result.
Changing the model can change behavior and cost. Confirm the intended configuration before restarting important work.
## Pricing and purchases [#pricing-and-purchases]
Use the price and billing information currently shown by the product or provider. This guide does not promise a free tier, unlimited use, or a fixed provider price.
For help identifying a blocked path, see [Agent not replying](/docs/help/agent-not-replying).
# Workspaces and members (/docs/team/workspaces)
## Choose a workspace [#choose-a-workspace]
Use the workspace switcher to check where you are before adding files, connecting tools, or creating work.
A **Personal Workspace** is for you. An **Organization** is for shared work with members. Creating an Organization does not move your personal files, conversations, or connected tools into it.
## Create or join an Organization [#create-or-join-an-organization]
Use the workspace switcher or account setup to create an Organization, join with an invitation, or select an existing membership.
Choose a recognizable name. If your team already has an Organization, ask for an invitation instead of creating a duplicate.
## Invite teammates [#invite-teammates]
Owners and admins can manage invitations.
1. Open the Organization's member or invitation settings.
2. Enter the intended email addresses, or create a join link if that option is available.
3. Check the Organization and role.
4. Send the invitation and inspect its status.
An invitation sent is not a member joined. Check membership after acceptance.
Share join links only with their intended audience. Review or revoke unused invitations through their management controls.
## Roles [#roles]
Owners and admins have management capabilities that ordinary members do not. Specific features can add narrower roles, such as [Skill Maintainer](/docs/team/skills).
If a control is missing, first check the active Organization and your role. Do not use another person's account to work around a restriction.
## Team versus Space access [#team-versus-space-access]
Being in the same Organization does not grant access to every private Space. Add people to the appropriate conversation deliberately.
Before removing a member, review their ongoing work and shared assets. Removing access and deleting shared work are different actions.
Next: [Access and sharing](/docs/team/access).
# Understand Agent Activity (/docs/work/activity)
Agent Activity helps your team see which local agent sessions are active, their Project, and when they were last observed.
It is an awareness view. It is not a work reservation, an assignment, or proof that an Agent completed a request.
## Enable observation deliberately [#enable-observation-deliberately]
Local observation is opt-in and limited to confirmed [Project folder connections](/docs/work/projects). Check the computer and Project before enabling the relevant Activity settings.
A folder found during discovery is not an approved connection.
## Open Activity [#open-activity]
Open a Project's details and use its related **Agent Activity** link. Activity may also be available through workspace navigation or search; it is not always in the primary rail.
Filter or group the view to find the Project and Agent you need. Check the last observation time before treating a status as current.
## What can appear [#what-can-appear]
Activity can include the agent tool, its user-visible session title, model, state, branch, and timestamps. Supported details depend on the tool.
Session titles can themselves contain meaningful information. Review your sharing choice even though Activity is not a transcript view.
## What it does not share [#what-it-does-not-share]
Activity does not expose full prompts, replies, reasoning, tool output, file contents, diffs, terminal commands, credentials, or absolute local paths.
Removing a Project connection stops new collection for that folder. It does not rewrite already accepted history.
## Before you act on an observation [#before-you-act-on-an-observation]
Confirm responsibility with the teammate doing the work. Two agents can be active on related work without either owning the whole task.
To check an OpenMeld request, open its original Space, thread, or Routine Run. Use the result and failure details there.
Next: [Track Space Tasks](/docs/work/tasks).
# Create and use an Agent (/docs/work/agents)
An Agent Profile is the named teammate you see in conversations. Its settings choose how and where it works.
## Choose Local or Remote [#choose-local-or-remote]
| Type | What you need | Where work runs |
| ------------ | ------------------------------------------------------------------------------ | -------------------------------- |
| Local | A connected computer, an available agent tool, and the intended working folder | On that computer |
| Remote | An available hosted model and permission to create the Agent | In OpenMeld's hosted environment |
| Center Agent | Your workspace's managed Center Agent | Hosted by OpenMeld |
Remote does not mean the Agent can automatically read files from your laptop. [Share the required files](/docs/work/files) or connect the required tool.
## Create an Agent [#create-an-agent]
1. Open **Team → Agents**.
2. Use the creation action and choose **Local** or **Remote**.
3. For Local, select the computer and agent tool. Complete any connection or sign-in action shown.
4. Select the available model and settings.
5. Check the proposed name, or enter a name that helps your team recognize the Agent.
6. Save and inspect its current availability.
For a Local Agent, check the working folder or Default Project before sending a task that reads or changes files. A folder path on one computer is not a folder on another.
## Give it work [#give-it-work]
Open a direct message or add the Agent to a [Space](/docs/work/spaces). Send a clear request. In a shared conversation, select the intended Agent and keep the addressing mode set to ask it to work.
**Mention only** adds a reference without starting work.
## Change settings carefully [#change-settings-carefully]
The Agent's owner manages its execution settings. Model and permission choices depend on the selected tool and available account. A more permissive mode can allow more actions; choose it deliberately.
Changing the Agent's name does not change the computer or tool that runs it. After changing execution settings, check availability again.
## When it cannot reply [#when-it-cannot-reply]
Use the reason shown on that Agent. Local recovery may involve its computer, tool sign-in, or folder. A hosted Agent's failure is not fixed by connecting your laptop.
See [Agent not replying](/docs/help/agent-not-replying).
# Work with Center Agent (/docs/work/center-agent)
Center Agent is the hosted AI teammate for your current workspace. It can answer questions, help organize work, and use the capabilities available to the person making the request.
## Start a conversation [#start-a-conversation]
1. Open **Center Agent** from the bottom toolbar.
2. Open the full chat if you need its history, files, or other views.
3. Write the result you want. Add the information needed to produce it.
4. Read the reply, check the result, and continue in the same conversation.
Start a new conversation for unrelated work. Switching conversations changes the discussion context; it does not create another Center Agent.
## Write a useful request [#write-a-useful-request]
Include the source, output, and limits. For example:
> Use the notes below to draft a customer update. Keep it under 150 words. Separate shipped changes from planned work. Save the draft here; do not email it.
For work in another app, name the account or repository and what the Agent may change. If a required connection is missing, [connect the Plugin](/docs/team/plugins) before expecting that action to work.
## Follow work and respond [#follow-work-and-respond]
Read progress and any request for information or approval. A question from the Agent may be the next required step.
Use the conversation's available stop control when you want work to stop. Check the resulting state. A click alone does not prove an external operation was canceled.
## Review the outcome [#review-the-outcome]
Check that a claimed file exists, a draft contains the right facts, or a change appears at its stated destination. Ask for the source when a factual claim matters.
Use **Files** for artifacts attached to the conversation. **Graph** offers a visual view of the conversation's work. Neither replaces the actual result.
## Suggestions and memory [#suggestions-and-memory]
Center Agent may suggest an improvement during a conversation. You can accept, defer, or dismiss it. Accepting a suggestion does not, by itself, authorize every action needed to implement it.
[Memory](/docs/work/memory), [Skills](/docs/team/skills), and [shared knowledge](/docs/team/knowledge) supply different kinds of reusable context.
If an owner or admin pauses Center Agent, work may be unavailable until it is resumed in Center Agent settings.
# Use a cloud Computer (/docs/work/cloud-computer)
A cloud Computer is a personal execution environment that an authorized hosted Agent can use for file and command work. It is separate from a [connected local computer](/docs/start/connect-computer).
## Ask for a bounded result [#ask-for-a-bounded-result]
In a Center Agent conversation, state the input, task, and output. For example:
> Create a CSV with the project names and dates from these notes. Save it and give me the resulting file. Do not send it to another service.
The Agent must have access to the source material and a supported execution path. A cloud Computer does not automatically contain your laptop's files, browser sessions, or external accounts.
## Check the actual outcome [#check-the-actual-outcome]
For file work, ask for the saved file and inspect it.
For code work, ask the Agent to identify the checkout, commands it ran, and results of the relevant checks. Access to a repository is not the same as a complete checkout. A checkout is not the same as installed dependencies or a working build.
If the Agent cannot complete a step, it should identify that step and the remaining work.
## Persistence [#persistence]
Saved Computer files can persist across runtime recreation. Running processes, installed operating-system packages, and browser login state are not promised to survive it.
Keep important deliverables in the appropriate [Files location](/docs/work/files). Review the audience when publishing an artifact from a personal execution environment into a team Space.
## Separate simultaneous work [#separate-simultaneous-work]
When tasks edit the same repository or files, ask for separate working copies or separate output files. Concurrent work does not automatically mean isolated work.
## Access and sharing [#access-and-sharing]
The Computer belongs to a human account. Access to its files does not authorize publishing everything in it.
Connected Plugins retain their own permissions. Removing a provider connection stops new provider calls; it does not automatically delete every file previously downloaded from that provider.
Cloud Computer capabilities do not imply a human remote-desktop app or a persistent signed-in browser.
# Share and find files (/docs/work/files)
## Choose a location [#choose-a-location]
| Location | Who can normally access it |
| ------------------------ | ------------------------------------- |
| Team files | Current Organization members |
| My files | You, within the current Organization |
| Space files | Authorized participants in that Space |
| Personal Workspace files | You in that Personal Workspace |
An Organization admin does not gain ordinary access to your **My files** simply by being an admin.
## Upload and open a file [#upload-and-open-a-file]
1. Open **Team → Files**, or **Files** in the target Space.
2. Check the location and audience.
3. Open the intended folder.
4. Drop a file into the file browser or use its upload control.
5. Wait for upload completion, then open the file.
An upload in progress is not a saved file. If it fails, use the visible retry or error guidance.
Preview support and Agent reading support are different. A file opening in a viewer does not mean every Agent can process that format.
## Give an Agent a file [#give-an-agent-a-file]
Attach or reference the file in the relevant conversation. State what the Agent should do with it and where the output should go.
For example:
> Use this spreadsheet to summarize the three largest changes. Post the summary here and identify the rows you used.
A private file may require an execution environment that can keep it private. A one-to-one chat does not automatically make every hosted environment suitable.
## Copy to another location [#copy-to-another-location]
Use **Copy to Space** or **Copy to Team**, and check the destination audience before confirming.
The copy is a separate file with its own history. Later source edits do not update it. Removing the source does not remove every independent copy.
## History and recovery [#history-and-recovery]
Use the file's available history controls to inspect revisions. If another edit conflicts with yours, review the latest version before saving again.
Recover ordinary deleted files through **Trash** where available. Managed **Team Knowledge** has its own editing and approval rules.
Public file links, automatic desktop folder sync, and automatic cross-Organization sharing are not part of this Files workflow.
Next: [Shared knowledge](/docs/team/knowledge).
# Understand and manage memory (/docs/work/memory)
Memory helps Center Agent retain useful context across conversations. It is separate from the messages and files that record the work itself.
## Member and Organization memory [#member-and-organization-memory]
**Member memory** belongs to you within one Organization. Eligible private Center Agent conversations can contribute to it.
**Organization memory** is shared workspace context. It is enabled by default unless changed. Shared conversations do not automatically become saved memory; explicit shared facts follow the Organization's memory policy.
A preference stated in a shared conversation is not automatically a private saved preference.
## Inspect the controls [#inspect-the-controls]
Open Center Agent and find **Memory** in its navigation or menu. Review the scope before changing a memory setting or clearing memory.
Clearing your member memory does not delete the conversation, its files, or shared Organization memory. Deleting a message is also not the same operation as clearing memory.
## Give durable instructions the right home [#give-durable-instructions-the-right-home]
Use [Skills](/docs/team/skills) for repeatable procedures. Use [Team Knowledge](/docs/team/knowledge) for approved reference information.
For important work, name the current source and ask the Agent to read it. Do not assume an old remembered fact is still correct.
## Be explicit about audience [#be-explicit-about-audience]
Before asking the Agent to retain information, decide whether it is personal or shared. Do not place private details in a team conversation just to make them easier to remember.
A connected messaging account and current access rules still apply when you talk through Slack, Lark, or Feishu.
## Check a claim about memory [#check-a-claim-about-memory]
If the Agent says it remembers something, ask which source or context it used when that matters. A statement in a reply is not a substitute for checking a saved record or a current policy.
Next: [Access and sharing](/docs/team/access).
# Messages, threads, and mentions (/docs/work/messages)
## Send a message [#send-a-message]
Open the intended Space or direct message. Write your request, add the relevant files, and send it. Check the conversation and audience before sending.
A useful request names an outcome:
> Review this draft for unclear instructions. Suggest edits here. Do not change the original file yet.
## Ask an Agent to work [#ask-an-agent-to-work]
In a shared Space, select the Agent through the mention picker. Check the relation shown on its token:
* A **Wake** asks the Agent to work.
* **Mention only** refers to the Agent without asking it to run.
* Mentioning a person draws their attention; it does not start agent execution.
If you only want to discuss an unavailable Agent, **Mention only** can let you continue without trying to run it.
## Reply in a thread [#reply-in-a-thread]
Use a message's reply action to keep a focused discussion attached to that message. Continue there when the work is about the same request.
A thread gives readers context. It does not grant access to someone outside the Space.
## Edit, recall, react, and pin [#edit-recall-react-and-pin]
Use the message's action menu for the actions available to you. Editing or recalling a message does not necessarily undo work already performed in another system.
Use a reaction for a lightweight acknowledgement. Pin messages that the Space will need repeatedly.
If you need an Agent to stop an active task, use the work's stop control. Do not rely on deleting or editing the request to reverse an external change.
## Answer a question or approval request [#answer-a-question-or-approval-request]
Respond on the original work item or conversation. Check the exact action and scope before approving. An old approval for a different request is not interchangeable.
[Inbox](/docs/help/notifications) helps you find work waiting for your input.
## Find a conversation [#find-a-conversation]
Use the conversation list's search and filters, or **Search or run a command** in the app navigation. Open a message search result to return to its source.
If a result is unavailable, your access may have changed. Ask the Space owner to check membership rather than sharing another person's session.
# Find your way around (/docs/work/navigation)
## Three places to start [#three-places-to-start]
* **Inbox** shows messages, reviews, and work that needs your attention.
* **Spaces** holds shared conversations and direct messages.
* **Team** holds people, Agents, Plugins, Skills, and Files.
Check the workspace switcher before you start. Each Organization has its own membership and resources.
## Quick controls [#quick-controls]
The bottom toolbar gives quick access to the current **Computer**, **Center Agent**, and **Inbox**. Open a quick view for a short check or use its full-page action for more room.
The computer shown there is a named connection. Check that name before assigning local work.
Use **Search or run a command** to find available navigation actions, including **Computers**. Search can also return messages you are allowed to read.
## Sidebars and conversation views [#sidebars-and-conversation-views]
Collapse a sidebar when you need more reading space. Use a conversation's **Messages**, **Graph**, and **Files** views for the same selected conversation.
**Graph** is a visual way to inspect related conversation work. Returning to Messages takes you back to the conversation. Graph positions do not change who can read or act in it.
## Desk [#desk]
Where the Desk entry is available, it presents Space windows on a canvas. You can move and resize windows while continuing to use the ordinary Space inside each one.
Desk layout is a browser presentation choice. Moving a window does not move messages to another Space or change membership.
A Welcome Space may guide you to **Add your first agent**. Review the selected members, send the prepared introduction when ready, and read the reply. Adding members does not send the draft automatically.
## Settings [#settings]
Use the Settings control for account, appearance, notification, privacy, and workspace configuration. Available sections depend on your role and client.
For specific entry points, use the [feature guide](/docs/reference/feature-guide).
# Connect a Project folder (/docs/work/projects)
A Project identifies work shared by your Organization. Each computer can connect its own folder or checkout to that Project.
Creating a Project does not clone a repository, install dependencies, or make a local folder ready to run.
## Add a Project [#add-a-project]
1. Open a Local Agent's working-folder control and choose **From a project**.
2. Choose **Create project**, or open **Project details** for an existing Project. The Projects management page also offers **Add a Project**.
3. Enter the name and repository details requested by the form.
4. Save, then open **View details**.
Two Projects can have the same display name. Check the slug and repository to distinguish them.
## Connect a folder [#connect-a-folder]
1. Connect the computer where the folder exists.
2. Open the Project's details.
3. Under **This Computer**, choose **Connect folder**.
4. Choose a folder on the named computer, select a discovered match, or enter its path.
5. Choose **Check and connect**.
6. Confirm that the saved connection points to the intended folder.
Selecting a folder in the picker is not the final connection. A discovered match also needs confirmation.
## Choose a default [#choose-a-default]
If you have several clones or worktrees, add each connection deliberately. Use **Make default** for the folder OpenMeld should use by default on that computer.
Check the Local Agent's Default Project and working folder before assigning work. A Project name does not resolve a missing or moved folder.
## When the folder cannot be found [#when-the-folder-cannot-be-found]
If you moved or removed the folder, update the connection. If verification fails, read the error: an unavailable computer, a Git problem, and “no matching folders” are different conditions.
Removing a connection stops using that connection. It is not a command to delete the repository from your computer.
## Agent Activity [#agent-activity]
Project connections also define where opt-in [Agent Activity](/docs/work/activity) may observe local sessions. A suggested folder is not permission to collect activity.
# Schedule a Routine (/docs/work/routines)
A Routine is saved, scheduled work. A Run records an occurrence and its outcome.
Use a Routine for a task whose inputs, permission, and destination are clear. For example: “Each Monday, summarize the release notes and post the summary in our release Space.”
## Create a Routine [#create-a-routine]
1. Open **Routines**. In the current app this is available from Center Agent; newer workspace navigation places it under **Team**.
2. Choose the create action.
3. Write the instructions and expected output.
4. Choose the Agent if the editor offers an executor selector.
5. Set the schedule and check the timezone.
6. Choose where the result should appear.
7. Review and save.
Use only destinations you can access. If your editor offers Personal and Team visibility, remember that visibility and output destination are separate: personal instructions can still publish to a shared Space you select.
## Make the instructions self-contained [#make-the-instructions-self-contained]
Specify the source, reporting period, output, and permitted actions. For example:
> Every Monday, read the release notes from the previous seven days. Post a short summary in Release Planning. Link each source. If nothing shipped, say so. Do not create issues or change releases.
Scheduled work may run while you are away. A task that needs an interactive sign-in or new approval can fail or need attention.
## Check Run history [#check-run-history]
Open the Routine to review its history. Use **View result** or the completed result link to open the published answer.
Check failed Runs for the reason. A **skipped** Run means the occurrence did not execute; it is not a successful result.
## Pause, change, or stop [#pause-change-or-stop]
Pause prevents future scheduling. It does not prove that a Run already in progress stopped. Use the running work's controls when a stop is needed.
After editing, check the schedule, timezone, Agent, and destination again. Do not assume an instruction change reruns a completed one-time Routine.
For a Local Agent Routine, its computer must be available when execution is attempted.
# Work in a Space (/docs/work/spaces)
A Space is a shared conversation for a goal. Use one when other people or agents need to follow the same work.
## Create a Space [#create-a-space]
1. Open **Spaces** and choose **New Space**.
2. Give the Space a clear name and describe its purpose.
3. Review its access and membership settings.
4. Add the people and agents who need to participate.
5. Send the first message with the goal and expected result.
For example:
> We are preparing Friday's release. Use this Space for the checklist, test results, and launch decision. First, help me turn the notes below into a checklist.
## Add context [#add-context]
Explain what is in scope, what is out of scope, and how you will judge the result. Attach the files participants need.
Use the Space's guide or brief for durable instructions. Pin an important message or file when people need to find it again.
Before adding someone, check whether the Space contains information they should see. See [Access and sharing](/docs/team/access).
## Ask the right Agent [#ask-the-right-agent]
Select the Agent in the message composer when you want it to work. Check whether the mention asks for work or is **Mention only**.
Adding an Agent to the member list does not start a task or make its computer available. Check the Agent's status and send an explicit request.
## Keep related work together [#keep-related-work-together]
* Use **Messages** for the conversation.
* Use [Threads](/docs/work/messages) for a focused follow-up.
* Use **Files** for shared attachments and artifacts.
* Use **Graph** to explore the conversation visually.
* Use [Tasks](/docs/work/tasks) for deliverables that need ownership, review, or later closure.
Return to the original conversation to review the outcome. Do not treat a status badge or activity timestamp as the deliverable.
## Direct messages [#direct-messages]
Use a direct message for a conversation with one teammate or Agent. Space Tasks belong to ordinary shared Spaces, not direct messages.
Next: [Messages and threads](/docs/work/messages).
# Track Space Tasks (/docs/work/tasks)
A Task records a commitment that needs ownership, progress, or review. Keep quick questions in Messages. Use Tasks when someone needs to return to a deliverable later.
Tasks belong to ordinary shared Spaces. They are not available in direct messages or private Center Agent conversations.
## Create a Task from a request [#create-a-task-from-a-request]
1. Open the request's message menu.
2. Choose **Add to Tasks**.
3. State the deliverable and completion conditions.
4. Choose an assignee and, when needed, an independent reviewer.
5. Save and keep discussion in the source message or thread.
For example, “Review the signup flow” is too broad. A better completion condition is “List each blocking issue, provide reproduction steps, and link the tested build.”
## Understand Task status [#understand-task-status]
| Status | Meaning |
| ----------- | -------------------------------------------- |
| To do | The commitment is recorded. |
| In progress | The assignee has started the Task workflow. |
| In review | A submitted result needs its reviewer. |
| Done | The stated completion boundary was accepted. |
| Canceled | The commitment is no longer being pursued. |
Assigning an Agent is not proof that the Agent has started executing. Check the actual work and result in the conversation.
## Submit and review [#submit-and-review]
The assignee submits a result. Without a reviewer, submission completes the Task against its conditions. With a reviewer, the Task enters **In review**.
Review the exact submitted result. Accept it or request concrete changes. A newer result needs a fresh review.
## Cancel or reopen [#cancel-or-reopen]
Use the Task's available actions. Canceling stops further Task transitions and may request a stop for linked agent work. Check whether that stop succeeded; cancellation does not undo changes already made.
Reopening is explicit. A linked GitHub issue reopening does not automatically reopen a completed Task.
## Linked issues [#linked-issues]
Tasks can retain links to GitHub issues. Check both the Task and issue before declaring the work closed. Completing the Task does not silently close its GitHub issue.