# Run things on the owner's own machine or home from a phone chat

> Run things on the owner's own machine or home from a phone chat: 5 companies fully do it, 9 with a limit, 116 carry the tag. Every listing in Personal assistants with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/personal-assistants.chat-remote-control/
- Markdown: https://www.anchorterminal.com/niches/personal-assistants.chat-remote-control/index.md (~2,550 tokens)
- Slim: https://www.anchorterminal.com/niches/personal-assistants.chat-remote-control/index.min.md (~330 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/personal-assistants.chat-remote-control/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Four or more companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 5 | 9 | 116 | 10 | 6 |

## What counts as done

The command runs on the owner's hardware and the result (file, status, screenshot) returns in the same chat thread within the session. Only the owner's sender ID can trigger it.

How agents ask for it: start the test loop and tell me every 5 iterations; turn the lights off; write the spec file while I walk the dog; summarise what WHOOP says about last night

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Hermes Agent (Nous Research, Inc.)**: A gateway on 20-plus platforms runs the terminal tool on the owner's machine, denies senders not allowlisted or paired by default, and /stop halts a run; acc-26 and acc-32 run it daily.
- **nanobot (HKUDS)**: Shell exec and file tools run on the gateway host and reply in the same chat; Telegram and Discord require pairing when allowFrom is omitted, and /stop interrupts work. Slack and Mattermost DMs are open until dm.policy is set.
- **NanoClaw (NanoCo, Inc.)**: Each session runs in a Docker container on the owner's machine with shell and files, replies through WhatsApp, Telegram, Signal or others, and unknown senders are dropped pending an owner approval card by default.
- **OpenClaw (OpenClaw Foundation)**: Shell, files and node commands run on the owner's gateway host and reply in the same chat; dmPolicy defaults to pairing so unknown senders are ignored. acc-43 drove Codex CLI from Telegram, acc-01 ran a Mac mini daily.
- **ZeroClaw (ZeroClaw Labs)**: Shell and file tools run on the owner's host with approval prompts delivered in the originating channel, an empty peer set denies every sender, and the OS sandbox is auto-detected.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **AstrBot (AstrBotDevs)**: Local Computer Use gives shell, Python and file tools reachable from Telegram, Discord or Slack; id_whitelist is empty by default and OS isolation is opt-in and may be unavailable.
- **CowAgent (zhayujie)**: Bash, file and browser tools run on the owner's machine from WeChat, Feishu, Telegram, Slack or Discord; the default is full-access, any group member who mentions the bot reaches the same tools, and issue #2998 reports files sent into a group.
- **Grok Bot (Anysphere, Inc.)**: Local execution runs commands on the member's own machine through the desktop app with per-command approval by default; the mobile app reaches the same Bot.
- **IronClaw (Jasnah, Inc. d/b/a NEAR AI)**: Shell commands with environment scrubbing and injection detection run from Telegram or Slack, and only paired Telegram users are answered; channels need a public HTTPS webhook and shell is not sandboxed.
- **Octop (Tencent Cloud)**: Terminal AI+ and browser run on the self-hosted process with per-user JWT isolation and chat channels, but shell and browser permissions need configuration and CVE-2026-108581 exposes provider keys to any signed-in user.
- **Dots (OpenAI)**: A connected computer runs local Work and Codex tasks the owner directs from ChatGPT, Slack, Teams or text, but only one computer per thread with the app open (acc-25), and the dot itself is hosted.
- **PicoClaw (Sipeed)**: Shell and files run on the owner's board and reply in Telegram, Discord or others. An empty allow_from accepts everyone (the DingTalk guide says so), remote exec defaults differ between docs and source, and 15 CVEs are unpatched.
- **QwenPaw (Qwenpaw PRIVATE LIMITED)**: Shell runs on the owner's host behind Tool Guard and an OS sandbox when governance picks one, replying in the chat. dm_policy and group_policy default to open, so anyone who can message the bot is accepted until set to allowlist.
- **Skydive (Create, Inc.)**: Agents run on their own cloud machines; the Mac desktop app grants local file and app access only after an approval card, and only from web chat or the CLI.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Hark Pro (Hark Labs, Inc.)**: hark.com's robots.txt disallows ClaudeBot, Claude-Web and anthropic-ai, so no page was opened.

- not this job: AI Job Search (Mads Lorentzen): Runs inside Claude Code on the owner's machine; no gateway or chat app receives commands.
- not this job: asaply (Asaply Inc.): Hosted service with no access to the owner's hardware.
- not this job: Asmi (Humint Labs, Inc.): Hosted; no access to the owner's machine.
- not this job: Atticus (An individual developer (GitHub SaarthurR)): Takes requests only from its own shortcut and typing box on the same Mac; no Telegram, WhatsApp or Discord.
- not this job: Boba (humans& ai, inc.): Hosted; no access to the owner's hardware.
- not this job: Boski (Boski, Inc.): Hosted agent; no access to the owner's hardware.
- not this job: Brea (Brea AI LLC): Hosted assistant with its own browser; no access to the owner's machine.
- not this job: Caddy (Caddy Technologies Inc.): Hosted; no access to the owner's hardware.
- not this job: Claude Duet (Anthropic, PBC): Works on its own cloud computer and cannot see the files on the owner's machine.
- not this job: DeepTutor (HKUDS): The exec tool runs CLI apps inside the tutor's sandbox for study tasks; it is not a gateway to the owner's machine.
- not this job: Del (Human Interface, Inc.): Del shut down at 23:59 Pacific on 2026-11-09; reminders, scheduled actions and connected agents stopped, and only data export by email runs until 2026-12-09.
- not this job: folk (Nozomio, Inc.): Each user gets a cloud computer; nothing reaches the owner's own hardware.
- not this job: GAIA (The Experience Company, Inc.): A hosted assistant over integrations; no shell or file tools for the owner's machine.
- not this job: Gemini Spark (Google LLC): Runs in Google's cloud; no access to the owner's machine.
- not this job: Haven (Hello Haven, Inc.): Hosted; no access to the owner's hardware.
- not this job: HireAlpha (HireAlpha): Hosted iMessage contacts; no access to the owner's hardware.
- not this job: Instinct (Spear Street Technology, Inc. (d/b/a Instinct)): Hosted; no access to the owner's hardware.
- not this job: LearnHarness (No legal entity found. Copyright Amos Blomqvist): No chat gateway; the local HTTP API serves its own window.
- not this job: Leon (No legal entity found. Copyright Louis Grenard): Shell, files and computer use exist, but requests arrive only through the web app on the same network; client auth is off in the sample config.
- not this job: Lindy (Crivello Corp (doing business as Lindy)): Hosted; no access to the owner's hardware.
- not this job: Lobby (Lobby OS, Inc.): Hosted app on desktop and mobile.
- not this job: Natura tinyNature (Natura Inc.): Hosted agent reached through messaging and earbuds.
- not this job: OpenDuet (Tawkit, Inc. (d/b/a CopilotKit)): Work runs in each Bot's Docker computer, not on the owner's machine, and the channels are the web app, Slack and Teams.
- not this job: OpenInstinct (Merit Systems, Inc.): Runs in the owner's Vercel account with a hosted browser, not on their own machine.
- not this job: Orbits (Nemus AI, Inc.): Hosted app on iOS and Android.
- not this job: Orchid (Zero Email, Inc.): Hosted; nothing runs on the owner's machine.
- not this job: Pally (Pally Technologies, Inc.): Hosted; the optional Cloud Mac is a vendor-run machine, not the owner's.
- not this job: Poke (The Interaction Company of California Inc.): Hosted; the inbound API only sends Poke a message, and nothing runs on the owner's machine.
- not this job: szn (Nomadic Futures, Inc.): Hosted.
- not this job: Tab (Terrasoft, Inc.): Hosted; its own computer and browser.
- not this job: Talorys (Roc Yu (rociiu) and contributors): Runs on Cloudflare Workers, with no shell or home access.
- not this job: Tasker AI (DAPI, Inc. (dba Tasker AI)): Hosted service; nothing runs on the owner's machine.
- not this job: Tomo (Mapo Labs, Inc.): Hosted, reached by iMessage and an iOS app.
- not this job: Town (Town.com, Inc.): Hosted; nothing runs on the owner's machine.
- not this job: Wajo (Fo) (Wajo, Inc.): Hosted platform; nothing runs on the owner's machine.
- not this job: Zapia (Brainlogic AI SAS): Hosted app on iOS, Android and web.
- not this job: Zo Computer (Zo Computer Company): Commands run on Zo's cloud Linux computer, not on the owner's own machine or home network.

## What stops agents finishing

- the default on most harnesses runs shell on the host with no prompt and accepts messages from anyone until allowFrom is set (acc-16, acc-11)
- the assistant competes with the owner's work when run on the main computer and dies when it sleeps (acc-02)

## Around it

Before: speech-to-text.voice-note. After: coding-agents.run-task, home-automation.device-control.

## Where this comes from

Weight 10 of 100 in Personal assistants: 6 of 40 accounts are self-hosted owners driving their own machine from chat (acc-01, acc-07, acc-16, acc-25, acc-43, acc-44); this is what OpenClaw, Hermes and NanoClaw users write about most, though it sits at the edge of the category.

As data: https://www.anchorterminal.com/api/v1/use-cases/personal-assistants.chat-remote-control.json, or the MCP tool get_use_case with id personal-assistants.chat-remote-control. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
