The Lingo Playground Families Voice Keyboard Spec

An AgentForge product · Built on the open Directive Shorthand spec

Lingo Keyboard

Lingo is AgentForge's implementation of Directive Shorthand — the open specification — and the keyboard is where it meets your thumbs. The chip strip, the stack builder, and the Lingo mic put the shorthand inside every app on your phone, so the text itself is correct at the source — and even an agent that has never heard of the standard receives perfect tokens.

Coming soon onGoogle Play

In the forge — coming soon to Google Play. Android first, forged alongside agenticdroid, AgentForge's phone-tooling venture. Tap surface first, dictation second — exactly as specified.

Explore the Lingo Read the standard

Concept Previews

Under your thumbs.

Three moments: the stack builder composing a token as you tap, the Lingo mic turning speech into canonical shorthand, and the keyboard at work inside an ordinary chat. Typing is the primary surface — dictation is the supporting convenience.

01 · Build the stack by tapping.
Chips compose under the axis rules — a stack that can't be built can't be mistyped.
02 · The button is the marker.
Known-context dictation: speak the spoken forms, canonical tokens land as text.
03 · In the wild, in any app.
The text is correct at the source — so the agent on the other end just understands.

Product visuals are concept previews of an app in development.

The Design

Built for the tap.

Primary · For typers

The chip strip + stack builder

A Lingo row above the keys: pick a family, tap chips, and the stack composes with the axis rules enforced by the UI — same-axis taps replace, illegal stacks are unrepresentable. A live preview shows the fastest form and its plain-English gloss before you commit.

Secondary · For talkers

The Lingo mic

Pressing the mic is the marker — the keyboard knows the next words are a command, so dictation runs the same registry parser at full generosity and inserts canonical typed tokens. Say “reply voice brief”; re:vo/br lands in the field.

Bridge · Both habits

Typed-mode conversion

Type the spoken forms from habit — reply voice brief as literal text — and the suggestion strip offers the canonical conversion. The autocorrect pattern: offered, never silently applied. Your text changes only when you tap.

Privacy

Yours stays yours.

Privacy, stated as design constraints

A keyboard can see everything typed into it — the most trusted surface on the phone. So the Lingo Keyboard's posture is architectural, not promotional:

  • No logging, no analytics, no network permission in the default build. The parser is local data and local code; nothing about your keystrokes leaves the device through the keyboard.
  • On-device recognition preferred for the Lingo mic — and where a device can't offer it, the keyboard says so plainly at first use instead of quietly downgrading to a cloud service.
  • Dark on password fields. On password and one-time-code fields the Lingo row and mic disable entirely — no chips, no conversion, no exceptions.
  • Auditable by construction. The parser and registry are small, deterministic, and published with the standard — “it only inserts what the registry says” is checkable, not asserted.

And on iPhone? Apple's keyboard extensions can't access the microphone, so the Lingo mic can't exist there. iOS gets Layer 1 — the spoken forms, honored by the agent itself. That's exactly why the standard's voice layer had to be complete without the keyboard.

The Forge Pass

Forged with the forge.

Lingo is AgentForge's product line for the open Directive Shorthand specification — the reference implementation, supercharged — and the keyboard ships as part of that family. Every AgentForge toolkit speaks Lingo; Forge Pass members get the keyboard when it lands on Google Play, alongside every toolkit in the forge.

One pass. Every toolkit. The Lingo included.

Get the Forge Pass Hear the voice layer