Migrate auth to magic link
P1Ship Squad
hands-free · privacy-first
Dictate the run. Pick where your voice goes.
Push-to-talk that writes straight into the prompt — with a three-mode capability chain. Stream in-browser, transcribe fully on-device with wasm Whisper, or fall back to the cloud. You choose, and the choice sticks.
voice · push-to-talk● rec
listening
prompt⇧Space to talk
open the pr-review canvas and run it
autolocalcloud
on-devicewhisper-tiny.en · audio never leaves the browser
Push-to-talk (⇧Space)On-device Whisper — audio never leavesAuto / local / cloud, your call
01
Three ways to listen, one button
Voice resolves the best path your browser can run: the native Web Speech API streams live interim text in Chrome, Edge, and Safari; true local Whisper transcribes on-device anywhere WebAssembly runs; and a cloud Whisper-1 fallback covers Firefox and in-app webviews. One mic button, the same push-to-talk gesture, whatever the engine underneath.
voice · push-to-talk● rec
listening
prompt⇧Space to talk
open the pr-review canvas and run it
autolocalcloud
on-devicewhisper-tiny.en · audio never leaves the browser
02
On-device by default — audio that never leaves
Local mode loads onnx-community/whisper-tiny.en into the browser via transformers.js and runs inference entirely on your machine. The model is a one-time ~145 MB download cached in browser storage; every transcription after that is instant and offline. No audio, no transcript, no round-trip — your voice stays on the device.
compliance · trust layeraudit-ready
agent.coderAgent run· trace · 8c4f21a12:04:11
u · danaRole changed· member · owner12:04:42
proxyPII redacted· 3 replacements12:05:09
u · renBYOK key rotated· anthropic12:06:30
pii proxy · standardscrubbing
in email dana@acme.io, card 4242 4242 4242 4242
out email <REDACTED:email>, card <REDACTED:credit_card>
anthropic…aF3kopenai…9Qx2
AES-256-GCM03
Your privacy preference, locked in
An inline chip cycles between auto, local, and cloud, persisted to localStorage. Auto picks the fastest available engine; local is a hard privacy lock that never silently falls through to the cloud; cloud is the explicit BYO-OpenAI path. Pick once and the setting follows you across sessions.
cost · this month84% of budget
$842 / $1,000
near cap84% spentsoft @ $750
by feature
canvas$184
evals$93
agent runner$28
adversary$7
last 30 days$842 total
04
Speak, see it land in the prompt
Hold ⇧Space in the composer or tap the mic and start talking — final and interim text stream straight into the textarea you're already writing in, so the run picks it up on submit. Every transcription is recorded in the compliance audit log as a usage event, while the raw audio and text are deliberately never stored.
missions · board4 lanes
Backlog2
Draft changelog entry
In progress1
Refactor billing webhook
Ship Squadrunning
Done1
Add OpenAPI spec
Docs CrewTrace
Archived1
Spike: vector search
@all ship the billing fixSend
Web Speech · local · cloud
Capability modes
whisper-tiny.en (fp32)
On-device model
None
Audio leaving the device in local mode
Part of one platform
Voice works hand in hand with Apps & Surfaces.
Speed plus trust — prove your agents got better this week.
Voice is one piece of Stackon, the observability-first workspace for teams running Claude and Codex. Start free and instrument your first run today.