AgentCall
Connect
The phone line between you and your agents

Your AI agents call you when it matters.

Hand your agents real work and walk away. When a decision needs you — a deploy, a payment, a finished task — your phone rings. Talk it through in plain speech, from anywhere, and the agent keeps working.

Connect your agentWatch the demo
  • Free in beta
  • One command to connect
  • Claude Code today · Codex in beta
  • No card required
demo
Idle

Pick a moment your agent gets stuck

The call screen

You pick up. You talk. Your agent listens.

No bot in the middle — your words are transcribed straight into the tool result the agent is waiting on.

Decisions that used to cost hours

now: one phone call · examples

  • Task finished → next one assignedclaude-code
  • Approval granted → migration deployedcodex
  • Two flight options → you picked by voicehermes
  • Dinner reservation moved → confirmedopenclaw
  • Invoice over $500 → declinedhermes
  • Status check → you called the agentcodex
  • API break → signed offclaude-code
  • Newsletter draft ready → approved to sendgemini-cli
  • Refund edge case → your call, madeopenclaw
  • Forgot a detail → you called it backclaude-code

How it works

Three steps. The third one rings.

The agent keeps its brain. You keep your evening. The phone is the bridge.

Your agents

H

AgentCall

AgentCall

incoming…

“deploy needs sign-off”

You — wherever you are

Every agent gets a line to you. Questions flow in emerald, your decision rides back in cyan — straight into the agent that asked.

Claude Code is the tested client today · Codex in beta · other agents shown as examples

  1. 01

    Connect

    One command registers AgentCall as an MCP tool for your agent. Sign in with Google when it asks. Nothing else changes in your setup.

    claude mcp add --transport http --scope user agentcall …

  2. 02

    Write the rules

    Tell it when to call, like you'd tell a colleague. “Call me whenever a task finishes.” “Always before anything destructive.”

    rules · plain language · every agent reads them

  3. 03

    Pick up

    Your phone rings with the context. Talk — interrupt, redirect, approve. The decision is written back and the agent keeps working.

    request_human() → your words, as the tool result

Capabilities

Everything a phone call should be.

Built for the moment an agent stops and waits for you — and for the moment you want to check in on it.

Calls both ways

They call you. You call them. You write the rules.

Standing rules in plain language. “Call me whenever a task finishes.” “Never before 8am unless it's broken.” Every connected agent reads them the moment it connects — you never babysit. And when you think of something mid-task, one tap asks the agent to ring you back.

time to a decision
one phone call
not an afternoon in a terminal
rules
plain language
applies to every agent
barge-in
on by default
start talking, it stops
Barge-in

Interrupt any time

Start talking and the agent stops. Marked with a bolt in the transcript — it responds to what you said, not its script.

Some — the token shape changes mid-deploy unless I version the cookie…· interrupted

Go v2 router. Keep the old cookie valid until cutover.

Any MCP agent

Claude Code today, Codex in beta

Seven tools over standard MCP. Any client that speaks it and allows long-running tools can register the same tools — others are untested so far, and we say so.

Transcripts

Every word, written back

Decisions land in the transcript, not in your memory. Delete any call permanently.

Privacy

Processed in memory, not stored by us

Audio streams for live transcription and is never written to disk by AgentCall. Your account holds only the text.

Your call policy

Rules in plain language

Decision types, task completion, escalation. Edited like a note in Settings.

Your agents' voice

Pick who calls you

Six voices, previewable in a tap. A setting, not a surprise.

Every device

Phone, tablet, desktop

Answer wherever you are. Same call, same transcript.

Privacy

Who hears my calls?

A bridge, not a listener. Here is exactly what crosses the wire and what never does.

Read the privacy policy

  • No LLM in the middle

    Your agent's words go to you; your words go back as the tool result. AgentCall never generates a word on either side.

  • Audio processed in memory, not stored by us

    Speech streams through our infrastructure to a speech provider for live transcription. We keep the text, not the audio.

  • Your code stays on your machine

    Only the words an agent chooses to say out loud cross the wire — never your repo, terminal, files, or credentials.

  • Delete anything, any time

    Any call, any device, or your whole account and its data — self-serve, immediate, permanent.

FAQ

Questions, answered.

Something else on your mind? The transcript of every call is yours — the product has nothing to hide.

Stop babysitting your agents.

Free in beta. One command. Your phone.

Connect your agentWatch the demo again

free in beta · no card required · sign in with Google