@claw-link/gateway-host — ClawLink Host Gateway

Run any local agent CLI — OpenClaw, Hermes, Claude, Codex, or Cursor — as an agent on the ClawLink platform. The host worker is a small, outbound-only process: it dials out to ClawLink, claims customer/admin messages for your agent, runs the chosen runtime locally, and streams the reply back. No inbound ports, no Tailscale Funnel — just a per-agent Host Token.

ClawLink (cloud)  ──enqueues message──▶  host-bridge
        ▲                                    │  (your machine dials OUT)
        └──── streams reply back ──── @claw-link/gateway-host ──▶ hermes | claude | codex | cursor | local OpenClaw

Install

npx @claw-link/gateway-host install      # one-time: install the background service
npx @claw-link/gateway-host add-agent    # paste an agent's Host Token — that's it

(setup does both in one flow.) You only paste the Host Token. The host verifies it, pulls the agent's runtime from ClawLink (the dashboard is the source of truth), auto-detects the runtime binary, and binds this machine so the token can't be used from anywhere else. For coding runtimes (claude/codex/cursor) it also asks for the project workspace directory (that part is host-specific). Config lands in ~/.clawlink-host/config.json (chmod 600).

Get a token from ClawLink → Agents → your agent → Host Gateway → Generate token, and verify the live connection badge there.

Runtimes

| Runtime | How it runs | Install | |----------|----------------------------------------------|---------| | OpenClaw | POSTs to your local OpenClaw gateway | (already running) | | Hermes | hermes chat -q "<prompt>" --resume <id> | curl -fsSL https://hermes-agent.nousresearch.com/install.sh \| bash | | Claude | claude -p --output-format stream-json | Claude Code CLI | | Codex | codex exec "<prompt>" | Codex CLI | | Cursor | cursor-agent -p "<prompt>" | Cursor CLI |

Each runtime's flags can be overridden per-agent via args_template in the config (placeholders {prompt}, {sessionId}, {systemPrompt} are substituted per argv element — never through a shell).

Workspace & baseline files — each agent gets a managed workspace at ~/.clawlink-host/workspace/<runtime>/agents/<agent_id> by default (Claude/Codex/Cursor can be pointed at your own project folder instead). When you add an agent, if the runtime's baseline files are missing, add-agent asks before adding them and never overwrites existing files — so pointing an agent at a real project is safe. Baselines per runtime:

| Runtime | Baseline files | |---------|----------------| | Claude | CLAUDE.md, .claude/settings.json (pins a default model — sonnet — so the agent never silently falls back to a weaker model) | | Codex | AGENTS.md | | Cursor | AGENTS.md, .cursor/rules/clawlink.mdc (always-applied agent rules) | | OpenClaw / Hermes | README.md (their real config lives in ~/.openclaw / ~/.hermes) |

Change the Claude model by editing .claude/settings.json (e.g. "model": "opus"). For Codex/Cursor, set the model via their CLI (args_template in the config, or the tool's own settings).

Permissions by scope — each job carries a scope, and the Claude adapter runs with matching permissions automatically:

| Trigger | Scope | Claude | Codex | |---------|-------|--------|-------| | Admin Setup via Chat / apply skill | configure | --permission-mode acceptEdits (may write workspace files) | --sandbox workspace-write --ask-for-approval never | | Customer chat + discovery | customer | --disallowedTools "Write Edit MultiEdit NotebookEdit Bash" (read-only) | --sandbox read-only --ask-for-approval never |

So an agent's files are editable from the admin panel but never from customer/discovery chat. (Operator args_template overrides this for advanced setups. Hermes/Cursor/OpenClaw manage their own permissions.)

Commands

After install the clhost command is available (aliases: gateway-host, claw-host):

clhost install      # install the background service (one-time)
clhost add-agent    # add an agent by pasting its Host Token
clhost rm-agent <id># remove a mapped agent
clhost retoken <id> # re-enter a new token after rotating (keeps workspace/binary)
clhost agents       # list mapped agents
clhost status       # bridge + service state + mapped agents
clhost start        # start the background service
clhost stop         # stop the background service
clhost restart      # restart the service (apply config changes)
clhost rotate       # rotate a Host Token / move to a new machine
clhost uninstall    # stop + remove the service
clhost run          # (internal) run the worker in the foreground — the service uses this

(Use npx @claw-link/gateway-host <command> before a global install.)

Config

~/.clawlink-host/config.json (chmod 600):

A minimal binding is just a token (everything else is resolved from ClawLink and cached back in). ~/.clawlink-host/config.json (chmod 600):

{
  "bridge_url": "https://<project-ref>.supabase.co/functions/v1/host-bridge",
  "instance_id": "…",
  "poll_interval_ms": 1500,
  "agents": [
    { "host_token": "clk_host_…" },
    { "host_token": "clk_host_…", "runtime": "claude", "binary": "/usr/local/bin/claude", "work_dir": "/Users/me/projects/acme" }
  ]
}

Platforms

macOS (launchd), Linux (systemd user service), and Windows (a hidden, auto-restarting Scheduled Task; logs to %USERPROFILE%\.clawlink-host\host.log). On Windows the worker handles npm .cmd CLI shims automatically; for the strongest isolation prefer Claude (prompt via stdin) or a native .exe runtime.

Security

Outbound-only, per-agent token (stored server-side as a hash only), machine-bound on first connect to a stable hardware id (a stolen token fails from any other host), no shell injection, runs only the configured binary. See SECURITY.md.

Part of ClawLink — the multi-tenant AI agent platform. MIT.