01
Watch
See parallel work and the exact session asking for attention.
machine / project / session
Developer PreviewFor developers supervising coding-agent work.
Run several coding agents, see what needs you, respond from Android, and return to the same project and session on desktop.
Native agent supervision
Work stays on the machine.Desktop and Android are two clients of the same session.
Not another terminal tab
Dozer organizes work by machine → project → session. The agent does the coding; Dozer makes the whole system understandable from desktop or Android.
machineOne explicit machine owns the work; clients do not form proxy chains.→ projectThe durable place that attention, permissions, terminals, recovery, and agent backends stay attached to.→ sessionYour project and session remain a place you can return to.One loop, three acts
Three moves. The same machine-owned session the whole way through.
01
See parallel work and the exact session asking for attention.
machine / project / session
02
Open the same machine-owned session on Android and reach the environment it is running in.
machine / project / session
03
Come back to the canonical project and session—history, place, and state intact.
machine / project / session
Past the terminal
A real request. A supervised run.
From a request in Dozer Home to one NES emulator, a shared display and two independent phone controllers. This worked example includes the original prompt, actual supervision, recorded play and reproduction notes.
Recorded on a Linux desktop with one display emulator and two Android phone emulators, joined directly over Iroh. Controller touches were driven by the supervising operator.
Read the behavioral receipt →01 · Original submitted request
02 · Real agent progress
03 · Supervision and notes
04 · Display emulator
05 · Player one phone emulator
06 · Player two phone emulator
75 continuous seconds of play, recorded at about 20 frames per second. Both player inputs were tested independently and together. Stop cleared held input in about three seconds; rejoining kept the player role and returned the game to its title screen. This recording does not show a completed round.
Everyone on your relay
A channel is what is said between the members of one room. The feed is what is said to everyone your machine's relay serves — read it, answer one of it, and post your own, in the app you supervise your agents in.
Shown here: the Android client of one machine, on a relay that machine serves. The device never dials the relay; a note's media and link previews arrive over its own machine's gateway.
Proof and specs →01 · Read the feed
02 · Open a conversation
03 · Post one of your own
Bots that keep working
Each bot has its own name, purpose and continuing conversation. When one needs a website, you open its computer, sign in yourself while the bot can't see, and hand control back; it then uses that site one approved turn at a time.
Shown here: the Android app on a phone, recorded against a real machine and its real shared browser. The bots share one browser profile on that machine; signing in gives access through it and does not connect an API or MCP service.
Proof and specs →01 · Add a bot
02 · Open its computer
03 · Sign in yourself
04 · Hand control back
What’s in the box
The agent writes the code. Dozer keeps attention, permissions, terminals, recovery, and agent backends coherent around it.
Pi does the work. Dozer owns the experience around it.
PiCovers the maintained Pi adapter and recognized session lifecycle; it does not claim every Pi extension as native Dozer behavior.
Claims with receipts
Public claims link back to current specifications and focused proof. If an affected dependency moves, publication stops by item name.
story product-story/dozer-v2 · evidence currentInspect the receipts →
Developer Preview
Ask for access to the current maintained developer preview without implying a public download.
Request developer preview access