Antivirus or Windows blocks the agent
Why it happens. On Windows, Local Commander uses three small helper programs (LocalCommanderAgent.exe, JobHost.exe and OwnerBridge.exe). They run commands and read files when your AI asks, which looks like what antivirus products are trained to catch. Some products, such as Kaspersky and Trellix, flag them on a general risk score, not because they found malware.
What you can do:
- Keep your protection on and don't restore files from quarantine. That's the right call.
- Check that the files are ours and unchanged. The security page has a short PowerShell snippet that compares your installed files with the published list. Every line should say "same". → /security#antivirus
- If Smart App Control is on (Windows 11), reinstall with the same install command. Since version 0.3.68 Local Commander switches to a mode without the helper programs. In that mode there are no screenshots or folder picker, and each command takes 1–5 seconds longer.
- On a managed corporate antivirus, ask your administrator for an exception for the single file by its SHA-256 hash, not for the whole folder.
Send us: the antivirus name and database version, the exact verdict, the file name and path, the time, and a screenshot of the quarantine entry with anything personal cropped out. Don't email full antivirus logs; they can contain personal data. Send those to your antivirus vendor and give us the case number.
What we do on our side. Every Windows release must pass Microsoft Defender and ClamAV on the exact package before it can be published. We submit our files to antivirus vendors for review, and keep the status of each one on the security page. We don't code-sign the helpers yet; instead we publish the fingerprint of every file so you can check exactly what you run.
The computer is online, but commands don't run
You see NATIVE_ADMISSION_PAUSED or ADMISSION_CAPACITY_BUSY, with nothing running.
Why it happens. This isn't a plan limit or the server. The agent on your computer pauses itself so nothing gets lost: its local command journal filled up, it lost access to that journal after sleep or because of antivirus, or it received a stop and is waiting for a restart.
What you can do:
- Restart only the agent, no reinstall: run
localcommander restart. Your pairing, permissions and files stay as they are. - If that doesn't help, update with the same command you installed with. On a computer that is already connected it updates in place; don't enter a new pairing code. Current versions clean the journal automatically and no longer pause because of it.
- To roll back, run the same command with the previous version number instead of
@latest.
Send us: the output of localcommander status. It shows which of the three causes it is.
Windows 10 (64-bit) is supported.
Your AI reports an error, but the command ran — or never started
An error in the chat doesn't mean anything happened to your files. Activity on localcmd.app shows what actually ran.
| What you see | What happened | What to do |
|---|---|---|
| TIMEOUT when fetching the result, although the command finished | Your computer's clock was more than 10 seconds fast. Our bug, fixed in agent 0.3.75 | Update the agent. Turn on "Set time automatically" and press "Sync now" |
| "Not started. Nothing ran" | Another command is holding the folder, usually a server started earlier. A new chat doesn't stop it. Activity now shows the real reason | Ask your AI to list running commands and stop the old one. Start long-running servers without a working folder (cwd) |
| tool_not_supported when creating a file | The drive uses exFAT (USB sticks, external drives) | Update to 0.3.68 or later |
Send us: the time of the attempt, the operation ID and the result of the diagnostics tool right after the error. If it keeps happening, turn on Full logs in the computer's details on localcmd.app, repeat one harmless attempt, download the file and send it to us. Turn Full logs off afterwards; that deletes the kept copies. Otherwise they are deleted after 7 days.
The computer won't connect
The agent says it started and then stops, and your AI sees no computers. Your network is fine.
Why it happens. Usually the computer was removed from your account on the website, often by accident. A removed computer's old pairing is refused for good, and current versions say so in plain words. On Windows 11 the other common cause is Smart App Control (see section 01).
What you can do:
- Open your computers on localcmd.app. If the computer isn't there, pair it again.
- Click "Add computer" and copy the code.
- In a terminal, clear the old pairing first (
uninstall --purge-state), then install with the code (--code YOUR_CODE). - Check
localcommander status: it should say running: true. - Refresh or reconnect Local Commander in your AI client.
Send us: the computer name, your account email and the output of localcommander status.
Your AI loses the connection to Local Commander
Connections used to drop now and then, and you had to sign in again. Since 16 November a sign-in stays valid for 30 days from last use and renews itself.
What you can do:
- Disconnect and reconnect Local Commander once in your AI client's connection settings.
- If your AI doesn't see new tools, refresh the connection's tools and start a new chat.
Send us: which AI client, roughly how long since the last sign-in, and the exact error text.
Quick answers
| Question | Answer |
|---|---|
| One long command, like a server, blocks everything else | Start long-running processes without a working folder (cwd), so the folder isn't locked |
| Antivirus damaged the install, and update or restart don't fix it | Run the install command again. It restores the files and keeps your pairing |
| The computer doesn't come back online after a reboot | Update the agent; this is fixed |
| How do I move a computer to another account? | Enter that account's "Add computer" code on it. The computer moves and its old connections are dropped |
| A large command output comes back cut | The chat gets the beginning and the end; your AI can fetch the full result separately |
| How do I connect a relative's computer when I'm far away? | You can approve the new computer from your own phone |
| "Path not allowed" although I gave full access | Most likely the path doesn't exist; current versions say "not found" |
Contact support from your AI
The quickest way: ask the chat where the problem happened. Your AI already has the operation IDs and error text, and it can collect your computer's state for us.
Option 1: let your AI send it. Say:
Something is wrong with Local Commander. Contact Local Commander support about it.
Your AI uses the support tool: it shows you the exact message first, including the details Local Commander adds (account, computer, agent version, recent errors). Nothing is sent until you say yes. We reply to your account email.
Option 2: from your account. Open the computer's details on localcmd.app and click "Contact support". The form fills in the computer and recent errors for you.
Option 3: by email. Ask your AI to draft it:
Local Commander isn't working right. Call the diagnostics tool for my computer and write an email to Local Commander support. Include the computer name, agent version, operating system, which AI I use, what I was trying to do, the exact error, the time with time zone and the operation IDs. Leave out passwords, keys, my file contents and full logs.
Send it to peter.mamrukov@localcmd.app.
What helps us:
| Detail | Why we need it | Where to find it |
|---|---|---|
| Account email | To find your account | localcmd.app |
| Computer name | Connection and command history | localcommander status or diagnostics |
| Agent version and OS | To see if it's already fixed | localcommander status |
| AI client | ChatGPT, Claude, Cursor and others behave differently | You know |
| Error code and time with time zone | To find the right record | The chat or Activity |
| Operation ID | To match it with our server | The chat |
| Diagnostics result | Agent state at the time of the error | Your AI calls it |
| For antivirus: name, verdict, screenshot | To submit the file to the vendor | Antivirus quarantine |
Don't send: passwords, keys or tokens, the contents of your work files, or full antivirus logs. Send Local Commander full logs only if we ask, and only after an attempt on a harmless folder.
Your dashboard
Five pages in the side menu: Computers, Activity, Usage, Plan and Settings. Below them, the computers your plan covers and your account.
Computers
Each computer and whether it is online now. Select ⋯ on a row to rename it, see why it is offline and how to bring it back, or remove it. Add computer shows one command for Windows, macOS or Linux; run it on that computer, sign in there with the same account and enter the code from the terminal.
Where you use it lists the AI apps you can connect: ChatGPT opens its directory listing; for Claude, terminal agents and any other MCP client it shows the address or command to copy.
Activity
Each card is one task your AI worked on, with what it did in everyday words. When your AI does not report where a task ends, the card covers its whole session. Open a card to see each request and its steps; failed ones open with the reason.
For each step you see the tool, the result, the time taken and the output size. Command text and output are not stored, on any plan, unless full logs are turned on for that computer. While the AI session is open (at most one hour) and the computer is online, your AI can fetch a step's output again. On the computer, localcommander monitor --json lists its most recent steps.
Usage
Usage counts executions: each step a request runs, so its numbers are larger than the number of requests. Only work sent through localcmd.app is counted, not everything your AI does with other tools. Periods are whole UTC days. Finished % is the share of executions that completed; not-started and unknown ones are counted apart, never as successes. Days without executions are not listed. CSV and JSON download the selected period.
Model tokens come from Claude Code and Codex: the token counts they send with your key, or their session files on computers where model token counts are on, from when each started reporting. AI apps in a browser or on a phone, other clients and billing are not measured. Task quality and dollar savings aren't measured yet. Local output-byte reduction is not a provider invoice. Missing data is shown as unknown, never as zero savings.
Settings
Your email, sign-in methods, preferred AI app and Sign out. AI apps with access lists each app that can send requests: access lasts 30 days from last use; select Keep access to extend it or Remove access to end it. Access lets your AI send requests. It doesn't mean a conversation is open or a computer is online. The clock icon shows open AI sessions. Screenshots (beta) are turned on here too.
Plan
Your plan, how many computers it covers and how many requests can run at once (parallel slots). Buying and changes go through PayPal on this website; AI apps never charge you.
AI agents and the Analyst
An agent identity counts only when a request carries project, task, agent-runtime, turn, operation and session links, with unknown shown as unknown. AI apps and sessions are connections, not identified agents. The agent and Analyst pages are hidden until they have real data for most accounts; AI-generated analysis is planned, not available yet.