v6: token input + WS connect + REST map - no browser auth needed

v5: Docker Playwright Google OAuth - UI password input, auto-save, auto-test

v4: app.sesame.com + stealth + human-like typing

fix: remove nest_asyncio, use thread for async

v3: Docker Playwright Google auth - one button token getter

v7: pure python webrtc aiortc

v6: capture WebRTC audio, Whisper transcription

v5: full WS dump, audio mute events, chat_get, raw message inspector

v4: fix gradio6 compat

v4: conversation capture, text messaging, fix null reply

v3: auto-reconnect on 30min limit, unlimited duration

FIXED: SDP embedded in call_connect, not standalone

Restore exact working auth from debug version

Step 1: memory bank + REST refresh, HF_TOKEN from secret

Step 1: memory bank + REST token refresh, token via env secret

Use Firebase REST refresh token - no browser auth after first run

Fix: increase auth timing to match working debug version

Format probe: try 4 offer content formats, add request_id/call_id nulls

Fix syntax error: missing except for inner try block

Auth debug: log every step of login flow

Fix: nest_asyncio to handle Gradio event loop conflict

Full debug: log ALL raw msgs, try 3 offer formats, show SDP codecs

Full debug: log ALL raw msgs, add call_id, call_start, relay policy

Full WebRTC negotiation: offer/answer/ICE in browser, fake mic, capture Maya audio track

Send text ping after connect, capture full response including audio frames

Fix: extra_headers -> additional_headers for websockets 14+

Follow Google redirect on main page, fill email+password there

Go directly to app.sesame.com/login

Navigate to /preview, dump page elements for debug

Fix: click Join the Beta first, then Google signin

Add app.py