Sculptural paper and metal streams converging around a coral signal cord

PRIVATE BETA · DISCORD

A meeting record.
With boundaries.

Toki joins when an authorized host asks, turns transient Discord voice into a speaker-attributed transcript, and keeps raw audio out of storage.

Private beta

Capture path

Discord Transient voice Final text Policy

Most meeting bots begin with the model.
Toki begins with the boundary.

Discord identity, lifecycle state, consent, transient media, durable text, and authorization each have a narrow job.

01 / CAPTURE

A meeting becomes a durable record.

Final English utterances keep their Discord speaker, meeting identity, timestamps, confidence, and provider metadata.

02 / CONTROL

A human decides when listening starts.

An authorized host starts Toki manually, the room sees a consent notice, and one server can run only one live meeting.

03 / TRUST

The model never owns permission state.

Read, disclose, and act remain separate fail-closed decisions, with workspace isolation and minimal audit records.

From “start the meeting”
to “show me the record.”

00:00

AUTHORIZE

A permitted host starts Toki

00:01

NOTICE

The linked text destination receives the consent notice

LIVE

PROCESS

Per-speaker audio stays transient through transcription

FINAL

PERSIST

Only finalized, attributed utterances become durable

LATER

READ

Authorized members request timestamped transcript pages

Every utterance knows where it came from.

Toki keeps the evidence needed to inspect a transcript without retaining the voice recording that produced it.

Discord speaker attribution
Meeting-relative timestamps
Confidence and provider metadata
No raw meeting audio persistence

Review the trust model

One focused stack.
Clear ownership.

The private beta uses narrow, replaceable boundaries instead of a connector marketplace.

Discord Transient media Deepgram STT PostgreSQL

Toki

Bring a trusted record
into your next call.

Private beta. Manual start. No raw audio stored.