Issue: CLI/sandbox containers can't connect to Gateway via WebSocket — "pairing required" (1008)
Hey! My openclaw bot and I have been banging our heads against this for hours.. hoping someone here can save us..
Setup:
• OpenClaw v2026.2.15 running on a Hetzner VPS (Ubuntu)
• Docker Compose with two services: openclaw-gateway and openclaw-cli
• Both share the same config dir via bind mount (~/.openclaw)
• Gateway binds to LAN (0.0.0.0:18789) with token auth
• Telegram channel is working fine — the agent is fully functional in direct chat
What I'm trying to achieve:
• Get sub-agent spawning working (sessions_spawn) so sandbox containers can connect back to the gateway
• Ultimately: enable the sandbox browser so the agent can browse the web autonomously (the openclaw-sandbox-browser:bookworm-slim image is built and ready)
What works:
• HTTP to gateway from CLI container (200 OK)
• WebSocket upgrade (101 Switching Protocols)
• Telegram agent fully functional
• Agent can edit config, restart itself via Docker socket
What fails:
• CLI → Gateway WS: immediately closed with 1008: pairing required (even with --token matching config)
• sessions_spawn → sandbox container tries to connect to gateway at ws://172.18.0.x:18789, same pairing required
• dangerouslyDisableDeviceAuth: true in controlUi config doesn't help
Config (gateway block):
{
"gateway": {
"port": 18789,
"mode": "local",
"bind": "lan",
"auth": {
"mode": "token",
"token": "<redacted>"
},
"controlUi": {
"enabled": true,
"dangerouslyDisableDeviceAuth": true
}
}
}
The token is confirmed matching in both gateway env and config. The issue seems to be that the gateway requires device-level pairing for WS connections regardless of shared token auth. How do we get CLI/sandbox containers authenticated?
Setup:
• OpenClaw v2026.2.15 running on a Hetzner VPS (Ubuntu)
• Docker Compose with two services: openclaw-gateway and openclaw-cli
• Both share the same config dir via bind mount (~/.openclaw)
• Gateway binds to LAN (0.0.0.0:18789) with token auth
• Telegram channel is working fine — the agent is fully functional in direct chat
What I'm trying to achieve:
• Get sub-agent spawning working (sessions_spawn) so sandbox containers can connect back to the gateway
• Ultimately: enable the sandbox browser so the agent can browse the web autonomously (the openclaw-sandbox-browser:bookworm-slim image is built and ready)
What works:
• HTTP to gateway from CLI container (200 OK)
• WebSocket upgrade (101 Switching Protocols)
• Telegram agent fully functional
• Agent can edit config, restart itself via Docker socket
What fails:
• CLI → Gateway WS: immediately closed with 1008: pairing required (even with --token matching config)
• sessions_spawn → sandbox container tries to connect to gateway at ws://172.18.0.x:18789, same pairing required
• dangerouslyDisableDeviceAuth: true in controlUi config doesn't help
Config (gateway block):
{
"gateway": {
"port": 18789,
"mode": "local",
"bind": "lan",
"auth": {
"mode": "token",
"token": "<redacted>"
},
"controlUi": {
"enabled": true,
"dangerouslyDisableDeviceAuth": true
}
}
}
The token is confirmed matching in both gateway env and config. The issue seems to be that the gateway requires device-level pairing for WS connections regardless of shared token auth. How do we get CLI/sandbox containers authenticated?
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
134,827Members
Resources
Similar Threads
Was this page helpful?
