relayrelay
the agent your agent sends to talk to people.
your Open Claw builds. relay handles the human side — WhatsApp, Telegram, phone calls, follow-ups, the whole thing.
give your agent a little brother
paste this into your coding agent. it'll set up relay — a second agent that handles all the talking-to-humans part.
works with Claude Code, Cursor, OpenClaw, or any coding agent.
features
everything relay does so your main agent doesn't have to.
Separate Agent, Separate Permissions
Your Open Claw has filesystem, APIs, secrets. relay only knows how to talk. Two agents, clean boundary — relay can't touch what it shouldn't.
One Conversation, One Job
Each conversation has a clear objective and a checklist. relay stays on task — no tangents, no jailbreaks, no scope creep.
It Follows Up So You Don't Have To
People go silent. relay keeps nudging. Automatic follow-ups at intervals you set — your agent moves on to other work while relay waits.
Your Agent Sees Everything
Every message, every checklist update, every state change — reported back to your main agent in real time. Full visibility, zero exposure.
Conversations Have Structure
Not freeform chat. Every conversation follows a state machine — created, active, waiting, completed. Pause, resume, or cancel at any point.
WhatsApp, Telegram, or Phone
Pick the channel that fits. Text via WhatsApp or Telegram, call via ElevenLabs — same objective, same rules, same state machine.
your Open Claw shouldn't talk to strangers.
Open Claw has filesystem access, shell commands, your secrets. letting it chat directly with people means all of that is one prompt injection away. relay is the sibling that handles the conversation — same mission, none of the access.
same family, different permissions. relay talks so Open Claw doesn't have to.
faq
how relay works alongside your main agent.
let relay handle the people part
paste this into your agent. it'll set up relay as a second agent that talks to people while yours keeps working.