shortcuts / brainstem-voice

🎙️ Brainstem Voice — retired design

No Shortcut, public endpoint, or install is shipped. Canonicalization, identity, frames, wire, eggs, registry, trust, and protocol evolution follow RAPP/1 rev-5 through RAPP1_AUTHORITY.json and RAPP1_STATUS.md.

Correct local design

The former Apple Shortcut idea is preserved as pre-acceptance design history. A future private experiment would:

  1. collect text;
  2. POST only required string user_input plus optional strings session_id and idempotency_key to http://127.0.0.1:7073/chat;
  3. read response from an exact three-member success, where agent_logs is an array;
  4. derive speech locally from that string; and
  5. speak the locally derived text.
{"user_input":"What is next?"}

{"response":"...","agent_logs":[],"session_id":"..."}

HTTP 422 is exactly {"error":{"code":"<code>","step":null}}.

Voice and twin stay inside the string

If response contains |||VOICE|||, a client may speak the following segment and stop before |||TWIN|||. Otherwise it may speak the full response.

There is no voice_response, twin_response, mode flag, client history, identity field, or other extra wire member.

Publication status

The promised iCloud link and downloadable .shortcut were never published. Tier 2 and public browser paths are retired. Do not present this page as an install surface or a conformant deployment.

Read the exact local adapter contract →