Have OpenClaw Call You on the Phone
Set up your agent to ring you—for a morning brief, a failure alert, or a quick check-in
Set up your agent to ring you—for a morning brief, a failure alert, or a quick check-in
You can already talk to OpenClaw with Voice Wake and Talk Mode. This guide flips it: your agent calls you—on your actual phone. Think morning wake-up brief, “your coding agent just failed” alert, or a nudge to log your gym session. Whenever you want the agent to reach out by voice, this is how.
Under the hood you’re combining ElevenAgents (from ElevenLabs) for the voice side and Twilio for the phone number. OpenClaw uses a skill to place the call through that stack.
How this fits with other voice features: Voice Wake and Talk Mode are for you talking to OpenClaw (wake word, then two-way speech). Some setups also let you call in to your bot. Here we’re only covering outbound—the agent initiates the call to your phone. Different use case, same idea: voice in the middle.
In this guide you’ll: install the ElevenLabs agents skill, add your API key and IDs, place your first outbound call, optionally tie calls to one agent or to a schedule, and know where to look when something breaks.
Two things have to be in place before we get to “call me”:
If that’s not set up yet, you’ll need to get ElevenAgents talking to OpenClaw and add Twilio in the ElevenAgents dashboard. Community guides and the providers’ own docs walk through that; here we’re only covering the outbound bit—getting your agent to dial you.
Pro tip: If the full setup feels like a lot, paste a step-by-step guide into your OpenClaw coding agent and ask it to do most of the work. It’s surprisingly good at that.
First, give OpenClaw the skills it needs to talk to the ElevenLabs API. In chat, ask your bot:
Install these skills globally: npx skills add elevenlabs/skills
That pulls in the skills that know how to use the ElevenLabs/Agents API.
Then add your ElevenLabs API key so the agent can actually authenticate:
openclaw dashboardWorth doing: create an API key with scope limited to agents-write and a sensible spend limit. Keeps things tighter if something goes sideways.
With the skill and API key in place, just ask your bot to call you. Something like:
Call <your phone number> using agents skill
The agent will use the key you added, then prompt you for two IDs:
Once a call works, tell OpenClaw to remember those so it doesn’t have to ask every time:
Put these IDs in your memory
It’ll stick them in memory. Next time you say “call me,” it can use the same Agent ID and Outbound Phone ID without bothering you.
Out of the box, each call might start a fresh OpenClaw agent. They all share memory, but not the same live conversation—so no “remember what we talked about last call” unless you do something extra.
If you want one agent that handles every call and keeps context between them, send a stable session id to your custom LLM using the x-openclaw-session-key header. Then the same logical agent answers every time.
Handy if you’re building a dedicated “calling” agent that should remember the last conversation or carry state from call to call.
To have the agent ring you automatically—say, 7am brief or when something needs your attention—use OpenClaw’s scheduling:
At the scheduled time the agent runs, pulls together the brief (or checks status, failures, whatever you asked for), and places the call using the agents skill and the IDs it saved. More on patterns like this in Automation and Example Setups & Cost.
When you set up cron (or ask the bot to set it up), you need to tell the agent what to do when it runs. Here are some concrete prompts you can adapt:
The agent uses the same “call me using agents skill” flow; the cron just decides when it runs and what context (brief, status check, etc.) to prepare before placing the call.
Common issues and what to check:
agents-write (or the scope your skill needs). Restart or re-open the dashboard and try again.For general OpenClaw issues, see Troubleshooting; for Gateway and channel problems, Setup Channel and provider docs (ElevenLabs, Twilio) are useful.
Outbound calls use two paid services: ElevenLabs (ElevenAgents) and Twilio. You’re charged per minute of voice and per phone segment (Twilio) and by usage on the ElevenLabs side. Rates depend on region and plan—check each provider’s dashboard and set spend limits or alerts so you’re not surprised. Storing the Agent ID and Outbound Phone ID in memory and reusing them doesn’t change cost; it just avoids re-entering them. If you schedule a lot of calls (e.g. many crons), keep an eye on usage.
When your agent places a call, the conversation is handled by ElevenLabs (voice) and Twilio (telephony). Your phone number and the call content pass through those providers according to their policies. OpenClaw only orchestrates the request (e.g. “call this number with this context”); it doesn’t store the audio. For best practices on API keys, scoped access, and data handling, see our Security guide and the providers’ own privacy and security docs.