Phone

Let an agent use your phone: texts, calls, speaker, microphone and apps

Pocket agents can now text from your number, place calls, talk out loud, listen, read your OTPs and open apps on your Android phone, with every risky action waiting for your tap.

Your agents already live on a server and report to your phone. Now it goes the other way: the phone is a tool the agent can use. Flip one switch and an agent running on your machine can send an SMS from your number, dial a call, switch the speaker on, say something out loud, listen through the microphone, read your recent texts and call log, look up a contact and open any app.

Turning it on

Account → Phone control → *Let agents use this phone*. Android asks for each permission separately, and the chips show exactly what you granted. Anything you deny simply is not offered to the agent: its tool list is built from your phone's real capabilities.

One switch in Account

What the agent may use

Tap the mic to dictate

What the agent can do

send_sms        text from your SIM (multipart, any length)
call / end_call dial a number, hang up
set_speaker     speakerphone on/off · set_volume
speak           text-to-speech out of the phone speaker
listen          speech-to-text from the microphone
read_sms        inbox/sent, filter by sender (OTPs, confirmations)
read_call_log   recent calls
find_contact    name → number from your contacts
list_apps       installed launchable apps
open_app        launch by name ("open Swiggy")
open_url        open a link or deep link
device_status   battery, charging, ringer, volume
notify          post a notification

Safe by default

• Ask mode can only read (status, contacts, apps). It cannot text, call or read your messages.
• Agent mode stops before every text and call and shows an approval card with the recipient and message. Reading texts or the call log asks once per conversation.
• Autonomous mode just does it. That's for routines you trust, like "every evening text me tomorrow's first meeting".
• Actions are queued on our server for three minutes, only your signed-in phone can claim them, and nothing about your SMS or contacts is stored beyond the one result the agent asked for.

Things people actually ask for

• "Read me the OTP that just came in and paste it into the login flow."
• "Call the clinic, put it on speaker, and tell me when someone picks up."
• "Text the team that the deploy is done", sent by the agent that did the deploy.
• "Every day at 9 PM, say my tomorrow's calendar out loud."

The composer also gained a mic: tap it, speak, and the words land in the message, so the whole loop from asking to acting can be hands-free.