chore: remove dead code (OpenAI voice backend + dead modules/functions)

.claude/rules/architecture.md

@@ -72,7 +72,7 @@
 
 Additional modules:
   screenshot.py       ─ Terminal text → PNG rendering (ANSI color, font fallback)
-  transcribe.py       ─ Voice-to-text transcription via whisper.cpp / Apple / OpenAI
+  transcribe.py       ─ Voice-to-text transcription via whisper.cpp / Apple Speech
   i18n.py             ─ Per-user UI strings (en / ru / zh)
   naming.py           ─ Haiku-generated session names + readable summaries
   usage.py            ─ Token usage aggregator + per-session token alerts
@@ -140,10 +140,10 @@ Handler modules (handlers/):
   quota_alerts.py     ─ Background /usage modal poll (default 10 min) →
                        5h/weekly band crossings 50/75/90 %
   inbox.py            ─ photo/document inbox under <workdir>/.ccbot-inbox/
-  interactive_ui.py   ─ AskUserQuestion / ExitPlanMode / Permission UI +
-                       adopt_interactive_msg / render_interactive_keyboard
-                       (used by switcher tap to claim the carrier as the
-                       interactive UI for a bg session whose prompt was stashed)
+  interactive_ui.py   ─ AskUserQuestion / ExitPlanMode / Permission UI
+                       (handle_interactive_ui + _build_interactive_keyboard).
+                       A switcher tap surfaces a bg session's stashed prompt
+                       via notifications.enter_kb_mode on the claimed carrier.
   directory_browser.py─ Directory + session picker UI builders
   switcher.py         ─ Inline session-switcher keyboard
   menu.py             ─ Footer / More / Settings keyboard composition;

.claude/rules/dm-architecture.md

@@ -116,7 +116,7 @@ When the user taps a session button in the main switcher:
 
 1. `transfer_card_to_carrier` pauses the FROM session's card and claims the carrier message_id for the TO session.
 2. `set_active_session(user, target)` flips the routing pointer.
-3. If the TO session has a stashed `bg_status.pending_interactive_ui` *and* the live pane still shows the prompt, the carrier is repainted with the prompt + the regular CB_ASK_* keyboard (`adopt_interactive_msg`). Otherwise the carrier is painted with `send_history` — the full paginated transcript view, with the standard footer ridden along as `extra_rows` so management controls stay reachable.
+3. If the TO session has a stashed `bg_status.pending_interactive_ui` *and* the live pane still shows the prompt, the carrier is claimed as the live card and flipped into kb-mode (`enter_kb_mode`) so the CB_ASK_* keyboard drives the prompt. Otherwise the carrier is painted as the session's live card (`paint_card_on_carrier` — header + paginated body + bg-panel + footer) and receives subsequent claude events in place.
 4. `bg_status.mark_seen` + `prune_seen` drop the just-viewed badge from the panel.
 
 Pagination (`CB_HISTORY_PREV/NEXT`) preserves the original `extra_rows` by stamping `context.user_data['_history_origin']` (`switcher` or `menu_list`) when the history view is first painted; the pagination handler rebuilds the matching footer from this hint. There is no explicit "History" button in the footer — pagination buttons themselves are the navigation affordance, and the user lands on the paginated view via switcher tap, Menu → List, or `/screenshot Back` (both `m` and `l` origins now paint history).

.claude/rules/secrets.md

@@ -10,7 +10,6 @@ where **not** to put) credentials when working on or with ccbot.
 | ccbot Telegram bot token | `~/.ccbot/.env` → `TELEGRAM_BOT_TOKEN=…` (or `./.env` in cwd) |
 | ccbot allowlist (Telegram user ids) | `~/.ccbot/.env` → `ALLOWED_USERS=…` |
 | ccbot outbound TG proxy (optional) | `~/.ccbot/.env` → `TG_PROXY_URL=…` |
-| OpenAI fallback voice key (optional) | `~/.ccbot/.env` → `OPENAI_API_KEY=…` |
 | Claude Code login token | `claude auth status` — managed by the CLI, not a file in the repo |
 | whisper.cpp model | `~/.ccbot/models/ggml-medium.bin` (path overridable via `WHISPER_MODEL_PATH`) |
 | ccbot persisted state | `~/.ccbot/state.json` — non-secret, but contains user ids / paths |

README.md

@@ -53,8 +53,7 @@ ways that are intentional and not negotiable:
   writes `session_map.json`; the monitor polls it. No reliance on
   process-tree introspection or claude SDK.
 - **Voice transcription is local-first.** `whisper.cpp` (default) or
-  Apple Speech via PyObjC on macOS. The OpenAI fallback exists but is
-  off by default — no API key required to run.
+  Apple Speech via PyObjC on macOS — no API key required to run.
 
 The full design rationale lives in `doc/dm-multisession-spec.md`. The
 implementation map is in `doc/dm-multisession-plan.md`.

README_CN.md

@@ -43,7 +43,7 @@ ccbot 让你可以:
 - **基于 hook 的会话跟踪。** Claude Code 的 `SessionStart` hook 写入
   `session_map.json`;监控器轮询它。不依赖进程树检查或 claude SDK。
 - **语音 — 本地优先。** `whisper.cpp`(默认)或 macOS 上通过 PyObjC
-  的 Apple Speech。OpenAI fallback 存在但默认关闭 — 运行不需要 API key。
+  的 Apple Speech — 运行不需要 API key。
 
 完整的设计动机在 `doc/dm-multisession-spec.md`。实现地图在
 `doc/dm-multisession-plan.md`。

README_RU.md

@@ -52,8 +52,7 @@ Claude Code живёт в терминале. Отошёл от стола — 
   пишет `session_map.json`; монитор бота его опрашивает. Никаких
   process-tree introspection или claude SDK.
 - **Голос — local-first.** `whisper.cpp` (по умолчанию) или Apple
-  Speech через PyObjC на macOS. OpenAI-fallback есть, но выключен —
-  API-ключ для запуска не нужен.
+  Speech через PyObjC на macOS — API-ключ для запуска не нужен.
 
 Полная архитектурная мотивация — в `doc/dm-multisession-spec.md`.
 Карта реализации — в `doc/dm-multisession-plan.md`.

doc/deploy.md

@@ -103,8 +103,6 @@ curl -s --max-time 8 -x "$TG_PROXY_URL" \
 - `VOICE_BACKEND=whisper` → requires `WHISPER_BIN` (default
   `whisper-cli`) and `WHISPER_MODEL_PATH` (default
   `$CCBOT_DIR/models/ggml-medium.bin`, ~1.5GB).
-- `VOICE_BACKEND=openai` → falls back to the legacy gpt-4o-transcribe
-  HTTP path; needs `OPENAI_API_KEY`.
 - `VOICE_BACKEND=off` → reject voice messages.
 
 ## State and disk usage

src/ccbot/bot/app.py

@@ -31,7 +31,6 @@
 from ..metrics import metrics_flush_loop
 from ..session import session_manager
 from ..session_monitor import NewMessage, SessionMonitor
-from ..transcribe import close_client as close_transcribe_client
 from ._common import CC_COMMANDS
 from .callbacks import callback_handler
 from .commands.info import (
@@ -90,7 +89,6 @@
 _status_poll_task: asyncio.Task[None] | None = None
 _card_timer_task: asyncio.Task[None] | None = None
 _quota_alerts_task: asyncio.Task[None] | None = None
-_context_poll_task: asyncio.Task[None] | None = None
 _metrics_flush_task: asyncio.Task[None] | None = None
 
 
@@ -101,7 +99,6 @@ async def post_init(application: "Application[Any, Any, Any, Any, Any, Any]") ->
         _status_poll_task, \
         _card_timer_task, \
         _quota_alerts_task, \
-        _context_poll_task, \
         _metrics_flush_task, \
         _conflict_app
 
@@ -173,15 +170,11 @@ async def message_callback(msg: NewMessage) -> None:
     _quota_alerts_task = asyncio.create_task(quota_alerts_loop(application.bot))
     logger.info("Quota alerts task started")
 
-    # Context-poll loop disabled — sending /context to live panes
-    # writes the modal's markdown output INTO the session's JSONL as
-    # a user-turn, which then renders on the live card as a fake
-    # ``[Request interrupted by user] ## Context Usage…`` block AND
-    # eats real tokens from claude's own context window every cycle.
-    # Context % is now computed from JSONL math (input + cache reads
-    # vs. 1M default for Claude 4.x models). See ``usage.context_pct_for_session``.
-    # _context_poll_task = asyncio.create_task(context_poll_loop(application.bot))
-    # logger.info("Context poll task started")
+    # Per-session context % is computed from JSONL math
+    # (usage.context_pct_for_session) — NOT by polling /context into panes.
+    # Polling wrote the modal's markdown into each session's JSONL as a fake
+    # user-turn (polluting the live card + burning tokens), so that path was
+    # removed. See doc/dm-multisession-spec.md §4.6.
 
     _metrics_flush_task = asyncio.create_task(metrics_flush_loop())
     logger.info("Metrics flush task started")
@@ -249,12 +242,7 @@ async def post_shutdown(
     application: "Application[Any, Any, Any, Any, Any, Any]",
 ) -> None:
     """Stop background tasks, flush queues, close HTTP clients."""
-    global \
-        _status_poll_task, \
-        _card_timer_task, \
-        _quota_alerts_task, \
-        _context_poll_task, \
-        _metrics_flush_task
+    global _status_poll_task, _card_timer_task, _quota_alerts_task, _metrics_flush_task
 
     if _status_poll_task:
         _status_poll_task.cancel()
@@ -283,15 +271,6 @@ async def post_shutdown(
         _quota_alerts_task = None
         logger.info("Quota alerts stopped")
 
-    if _context_poll_task:
-        _context_poll_task.cancel()
-        try:
-            await _context_poll_task
-        except asyncio.CancelledError:
-            pass
-        _context_poll_task = None
-        logger.info("Context poll stopped")
-
     if _metrics_flush_task:
         _metrics_flush_task.cancel()
         try:
@@ -315,8 +294,6 @@ async def post_shutdown(
         await session_monitor.stop()
         logger.info("Session monitor stopped")
 
-    await close_transcribe_client()
-
 
 def create_bot() -> "Application[Any, Any, Any, Any, Any, Any]":
     """Build the Application, wire all handlers, return it ready to run_polling."""

src/ccbot/bot/messages.py

@@ -607,14 +607,6 @@ async def voice_handler(update: Update, context: ContextTypes.DEFAULT_TYPE) -> N
     if config.voice_backend == "off":
         await safe_reply(update.message, "⚠ Voice is disabled (VOICE_BACKEND=off).")
         return
-    if config.voice_backend == "openai" and not config.openai_api_key:
-        await safe_reply(
-            update.message,
-            "⚠ VOICE_BACKEND=openai but OPENAI_API_KEY is unset.\n"
-            "Set the key or switch VOICE_BACKEND to whisper/auto.",
-        )
-        return
-
     wid = active_window(user.id)
     if wid is None:
         await safe_reply(

src/ccbot/config.py

@@ -131,12 +131,6 @@ def __init__(self) -> None:
             os.getenv("CCBOT_SHOW_HIDDEN_DIRS", "").lower() == "true"
         )
 
-        # OpenAI API for voice message transcription (optional)
-        self.openai_api_key: str = os.getenv("OPENAI_API_KEY", "")
-        self.openai_base_url: str = os.getenv(
-            "OPENAI_BASE_URL", "https://api.openai.com/v1"
-        )
-
         # --- DM multi-session mode ---
         # Sessions
         self.max_sessions: int = int(os.getenv("MAX_SESSIONS", "10"))

src/ccbot/handlers/__init__.py

@@ -8,5 +8,4 @@
   - directory_browser: Directory selection UI
   - interactive_ui: Interactive UI (AskUserQuestion, Permission Prompt, etc.)
   - status_polling: Terminal status line polling
-  - response_builder: Build paginated response messages
 """

src/ccbot/handlers/bg_status.py

@@ -97,11 +97,6 @@ def _touch(entry: BgStatus) -> None:
     entry.last_change = time.time()
 
 
-def has_panel_content(user_id: int) -> bool:
-    """True if the user has any bg-status entry worth rendering."""
-    return bool(_bg.get(user_id))
-
-
 def update_status(
     user_id: int,
     session_id: str,

src/ccbot/handlers/card_model.py

@@ -146,10 +146,6 @@ class Event:
     is_error: bool = False
     image_data: list[tuple[str, bytes]] | None = None  # tool_result images
 
-    @property
-    def is_tool(self) -> bool:
-        return self.type in ("tool_use", "tool_result")
-
 
 @dataclass
 class CardState: