feat: add telegram /model and /reasoning overrides (#147)

docs/reference/commands-and-directives.md

@@ -36,6 +36,8 @@ This line is parsed from replies and takes precedence over new directives.
 |---------|-------------|
 | `/cancel` | Reply to the progress message to stop the current run. |
 | `/agent` | Show/set the default agent for the current scope. |
+| `/model` | Show/set the model override for the current scope. |
+| `/reasoning` | Show/set the reasoning override for the current scope. |
 | `/file put <path>` | Upload a document into the repo/worktree (requires file transfer enabled). |
 | `/file get <path>` | Fetch a file or directory back into Telegram. |
 | `/topic <project> @branch` | Create/bind a topic (topics enabled). |

docs/reference/transports/telegram.md

@@ -58,7 +58,7 @@ Explicit invocation includes any of:
 - `@botname` mention in the message.
 - `/engine` or `/project_alias` as the first token.
 - Replying to a bot message.
-- Built-in or plugin slash commands (for example `/agent`, `/file`, `/trigger`).
+- Built-in or plugin slash commands (for example `/agent`, `/model`, `/reasoning`, `/file`, `/trigger`).
 
 Commands:
 

src/takopi/ids.py

@@ -7,7 +7,7 @@ _ID_RE = re.compile(ID_PATTERN)
 
 RESERVED_CLI_COMMANDS = frozenset({"init", "plugins", "doctor"})
 RESERVED_CHAT_COMMANDS = frozenset(
-    {"cancel", "file", "new", "agent", "trigger", "topic", "ctx"}
+    {"cancel", "file", "new", "agent", "model", "reasoning", "trigger", "topic", "ctx"}
 )
 RESERVED_ENGINE_IDS = RESERVED_CLI_COMMANDS | RESERVED_CHAT_COMMANDS
 RESERVED_COMMAND_IDS = RESERVED_CLI_COMMANDS | RESERVED_CHAT_COMMANDS

src/takopi/runners/claude.py

@@ -14,6 +14,7 @@ from ..events import EventFactory
 from ..logging import get_logger
 from ..model import Action, ActionKind, EngineId, ResumeToken, TakopiEvent
 from ..runner import JsonlSubprocessRunner, ResumeTokenMixin, Runner
+from .run_options import get_run_options
 from ..schemas import claude as claude_schema
 from .tool_actions import tool_input_path, tool_kind_and_title
 
@@ -296,11 +297,15 @@ class ClaudeRunner(ResumeTokenMixin, JsonlSubprocessRunner):
         return f"`claude --resume {token.value}`"
 
     def _build_args(self, prompt: str, resume: ResumeToken | None) -> list[str]:
+        run_options = get_run_options()
         args: list[str] = ["-p", "--output-format", "stream-json", "--verbose"]
         if resume is not None:
             args.extend(["--resume", resume.value])
-        if self.model is not None:
+        model = self.model
-            args.extend(["--model", str(self.model)])
+        if run_options is not None and run_options.model:
+            model = run_options.model
+        if model is not None:
+            args.extend(["--model", str(model)])
         allowed_tools = _coerce_comma_list(self.allowed_tools)
         if allowed_tools is not None:
             args.extend(["--allowedTools", allowed_tools])

src/takopi/runners/codex.py

@@ -13,6 +13,7 @@ from ..events import EventFactory
 from ..logging import get_logger
 from ..model import ActionPhase, EngineId, ResumeToken, TakopiEvent
 from ..runner import JsonlSubprocessRunner, ResumeTokenMixin, Runner
+from .run_options import get_run_options
 from ..schemas import codex as codex_schema
 from ..utils.paths import relativize_command
 
@@ -426,13 +427,26 @@ class CodexRunner(ResumeTokenMixin, JsonlSubprocessRunner):
         *,
         state: Any,
     ) -> list[str]:
-        args = [
+        run_options = get_run_options()
-            *self.extra_args,
+        args = [*self.extra_args]
+        if run_options is not None:
+            if run_options.model:
+                args.extend(["--model", str(run_options.model)])
+            if run_options.reasoning:
+                args.extend(
+                    [
+                        "-c",
+                        f"model_reasoning_effort={run_options.reasoning}",
+                    ]
+                )
+        args.extend(
+            [
                 "exec",
                 "--json",
                 "--skip-git-repo-check",
                 "--color=never",
             ]
+        )
         if resume:
             args.extend(["resume", resume.value, "-"])
         else:

src/takopi/runners/opencode.py

@@ -34,6 +34,7 @@ from ..model import (
     TakopiEvent,
 )
 from ..runner import JsonlSubprocessRunner, ResumeTokenMixin, Runner
+from .run_options import get_run_options
 from ..schemas import opencode as opencode_schema
 from ..utils.paths import relativize_path
 from .tool_actions import tool_input_path, tool_kind_and_title
@@ -325,11 +326,15 @@ class OpenCodeRunner(ResumeTokenMixin, JsonlSubprocessRunner):
         *,
         state: Any,
     ) -> list[str]:
+        run_options = get_run_options()
         args = ["run", "--format", "json"]
         if resume is not None:
             args.extend(["--session", resume.value])
-        if self.model is not None:
+        model = self.model
-            args.extend(["--model", str(self.model)])
+        if run_options is not None and run_options.model:
+            model = run_options.model
+        if model is not None:
+            args.extend(["--model", str(model)])
         args.extend(["--", prompt])
         return args
 

src/takopi/runners/pi.py

@@ -27,6 +27,7 @@ from ..model import (
     TakopiEvent,
 )
 from ..runner import JsonlSubprocessRunner, ResumeTokenMixin, Runner
+from .run_options import get_run_options
 from ..schemas import pi as pi_schema
 from ..utils.paths import get_run_base_dir
 from .tool_actions import tool_kind_and_title
@@ -322,11 +323,15 @@ class PiRunner(ResumeTokenMixin, JsonlSubprocessRunner):
         *,
         state: PiStreamState,
     ) -> list[str]:
+        run_options = get_run_options()
         args: list[str] = [*self.extra_args, "--print", "--mode", "json"]
         if self.provider:
             args.extend(["--provider", self.provider])
-        if self.model:
+        model = self.model
-            args.extend(["--model", self.model])
+        if run_options is not None and run_options.model:
+            model = run_options.model
+        if model:
+            args.extend(["--model", model])
         args.extend(["--session", state.resume.value])
         args.append(self._sanitize_prompt(prompt))
         return args

src/takopi/runners/run_options.py

@@ -0,0 +1,38 @@
+from __future__ import annotations
+
+from collections.abc import Iterator
+from contextlib import contextmanager
+from contextvars import ContextVar, Token
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True, slots=True)
+class EngineRunOptions:
+    model: str | None = None
+    reasoning: str | None = None
+
+
+_RUN_OPTIONS: ContextVar[EngineRunOptions | None] = ContextVar(
+    "takopi.engine_run_options", default=None
+)
+
+
+def get_run_options() -> EngineRunOptions | None:
+    return _RUN_OPTIONS.get()
+
+
+def set_run_options(options: EngineRunOptions | None) -> Token:
+    return _RUN_OPTIONS.set(options)
+
+
+def reset_run_options(token: Token) -> None:
+    _RUN_OPTIONS.reset(token)
+
+
+@contextmanager
+def apply_run_options(options: EngineRunOptions | None) -> Iterator[None]:
+    token = set_run_options(options)
+    try:
+        yield
+    finally:
+        reset_run_options(token)

src/takopi/telegram/chat_prefs.py

@@ -5,6 +5,7 @@ from pathlib import Path
 import msgspec
 
 from ..logging import get_logger
+from .engine_overrides import EngineOverrides, normalize_overrides
 from .state_store import JsonStateStore
 
 logger = get_logger(__name__)
@@ -16,6 +17,7 @@ STATE_FILENAME = "telegram_chat_prefs_state.json"
 class _ChatPrefs(msgspec.Struct, forbid_unknown_fields=False):
     default_engine: str | None = None
     trigger_mode: str | None = None
+    engine_overrides: dict[str, EngineOverrides] = msgspec.field(default_factory=dict)
 
 
 class _ChatPrefsState(msgspec.Struct, forbid_unknown_fields=False):
@@ -49,6 +51,13 @@ def _normalize_trigger_mode(value: str | None) -> str | None:
     return None
 
 
+def _normalize_engine_id(value: str | None) -> str | None:
+    if value is None:
+        return None
+    value = value.strip().lower()
+    return value or None
+
+
 def _new_state() -> _ChatPrefsState:
     return _ChatPrefsState(version=STATE_VERSION, chats={})
 
@@ -120,6 +129,45 @@ class ChatPrefsStore(JsonStateStore[_ChatPrefsState]):
     async def clear_trigger_mode(self, chat_id: int) -> None:
         await self.set_trigger_mode(chat_id, None)
 
+    async def get_engine_override(
+        self, chat_id: int, engine: str
+    ) -> EngineOverrides | None:
+        engine_key = _normalize_engine_id(engine)
+        if engine_key is None:
+            return None
+        async with self._lock:
+            self._reload_locked_if_needed()
+            chat = self._get_chat_locked(chat_id)
+            if chat is None:
+                return None
+            override = chat.engine_overrides.get(engine_key)
+            return normalize_overrides(override)
+
+    async def set_engine_override(
+        self, chat_id: int, engine: str, override: EngineOverrides | None
+    ) -> None:
+        engine_key = _normalize_engine_id(engine)
+        if engine_key is None:
+            return
+        normalized = normalize_overrides(override)
+        async with self._lock:
+            self._reload_locked_if_needed()
+            chat = self._get_chat_locked(chat_id)
+            if normalized is None:
+                if chat is None:
+                    return
+                chat.engine_overrides.pop(engine_key, None)
+                if self._chat_is_empty(chat):
+                    self._remove_chat_locked(chat_id)
+                self._save_locked()
+                return
+            chat = self._ensure_chat_locked(chat_id)
+            chat.engine_overrides[engine_key] = normalized
+            self._save_locked()
+
+    async def clear_engine_override(self, chat_id: int, engine: str) -> None:
+        await self.set_engine_override(chat_id, engine, None)
+
     def _get_chat_locked(self, chat_id: int) -> _ChatPrefs | None:
         return self._state.chats.get(_chat_key(chat_id))
 
@@ -136,8 +184,16 @@ class ChatPrefsStore(JsonStateStore[_ChatPrefsState]):
         return (
             _normalize_text(chat.default_engine) is None
             and _normalize_trigger_mode(chat.trigger_mode) is None
+            and not self._has_engine_overrides(chat.engine_overrides)
         )
 
+    @staticmethod
+    def _has_engine_overrides(overrides: dict[str, EngineOverrides]) -> bool:
+        for override in overrides.values():
+            if normalize_overrides(override) is not None:
+                return True
+        return False
+
     def _remove_chat_locked(self, chat_id: int) -> bool:
         key = _chat_key(chat_id)
         if key not in self._state.chats:

src/takopi/telegram/commands/dispatch.py

@@ -9,6 +9,7 @@ from ...commands import CommandContext, get_command
 from ...config import ConfigError
 from ...logging import get_logger
 from ...model import EngineId, ResumeToken
+from ...runners.run_options import EngineRunOptions
 from ...runner_bridge import RunningTasks
 from ...scheduler import ThreadScheduler
 from ...transport import MessageRef
@@ -33,6 +34,8 @@ async def _dispatch_command(
     on_thread_known: Callable[[ResumeToken, anyio.Event], Awaitable[None]] | None,
     stateful_mode: bool,
     default_engine_override: EngineId | None,
+    engine_overrides_resolver: Callable[[EngineId], Awaitable[EngineRunOptions | None]]
+    | None,
 ) -> None:
     allowlist = cfg.runtime.allowlist
     chat_id = msg.chat_id
@@ -52,6 +55,7 @@ async def _dispatch_command(
         running_tasks=running_tasks,
         scheduler=scheduler,
         on_thread_known=on_thread_known,
+        engine_overrides_resolver=engine_overrides_resolver,
         chat_id=chat_id,
         user_msg_id=user_msg_id,
         thread_id=msg.thread_id,

src/takopi/telegram/commands/executor.py

@@ -11,10 +11,11 @@ from ...commands import CommandExecutor, RunMode, RunRequest, RunResult
 from ...config import ConfigError
 from ...context import RunContext
 from ...logging import bind_run_context, clear_context, get_logger
-from ...model import EngineId, ResumeToken, TakopiEvent
+from ...model import Action, ActionEvent, EngineId, ResumeToken, TakopiEvent
 from ...progress import ProgressTracker
 from ...router import RunnerUnavailableError
 from ...runner import Runner
+from ...runners.run_options import EngineRunOptions, apply_run_options
 from ...runner_bridge import (
     ExecBridgeConfig,
     IncomingMessage as RunnerIncomingMessage,
@@ -26,6 +27,7 @@ from ...transport import MessageRef, RenderedMessage, SendOptions
 from ...transport_runtime import TransportRuntime
 from ...utils.paths import reset_run_base_dir, set_run_base_dir
 from ..bridge import send_plain
+from ..engine_overrides import supports_reasoning
 
 logger = get_logger(__name__)
 
@@ -53,6 +55,54 @@ class _ResumeLineProxy:
         return self.runner.run(prompt, resume)
 
 
+@dataclass(slots=True)
+class _PreludeRunner:
+    runner: Runner
+    prelude_events: Sequence[TakopiEvent]
+
+    @property
+    def engine(self) -> str:
+        return self.runner.engine
+
+    def is_resume_line(self, line: str) -> bool:
+        return self.runner.is_resume_line(line)
+
+    def format_resume(self, token: ResumeToken) -> str:
+        return self.runner.format_resume(token)
+
+    def extract_resume(self, text: str | None) -> ResumeToken | None:
+        return self.runner.extract_resume(text)
+
+    async def run(
+        self, prompt: str, resume: ResumeToken | None
+    ) -> AsyncIterator[TakopiEvent]:
+        for event in self.prelude_events:
+            yield event
+        async for event in self.runner.run(prompt, resume):
+            yield event
+
+
+def _reasoning_warning(
+    *, engine: str, run_options: EngineRunOptions | None
+) -> ActionEvent | None:
+    if run_options is None or not run_options.reasoning:
+        return None
+    if supports_reasoning(engine):
+        return None
+    message = f"reasoning override is not supported for `{engine}`; ignoring."
+    return ActionEvent(
+        engine=engine,
+        action=Action(
+            id=f"{engine}.override.reasoning",
+            kind="note",
+            title=message,
+            detail={},
+        ),
+        phase="completed",
+        ok=True,
+    )
+
+
 def _should_show_resume_line(
     *,
     show_resume_line: bool,
@@ -108,6 +158,7 @@ async def _run_engine(
     thread_id: int | None = None,
     show_resume_line: bool = True,
     progress_ref: MessageRef | None = None,
+    run_options: EngineRunOptions | None = None,
 ) -> None:
     reply = partial(
         send_plain,
@@ -128,6 +179,9 @@ async def _run_engine(
         runner: Runner = entry.runner
         if not show_resume_line:
             runner = cast(Runner, _ResumeLineProxy(runner))
+        warning = _reasoning_warning(engine=runner.engine, run_options=run_options)
+        if warning is not None:
+            runner = cast(Runner, _PreludeRunner(runner, [warning]))
         if not entry.available:
             reason = entry.issue or "engine unavailable"
             await _send_runner_unavailable(
@@ -167,6 +221,7 @@ async def _run_engine(
                 reply_to=reply_ref,
                 thread_id=thread_id,
             )
+            with apply_run_options(run_options):
                 await handle_message(
                     exec_cfg,
                     runner=runner,
@@ -235,6 +290,10 @@ class _TelegramCommandExecutor(CommandExecutor):
         running_tasks: RunningTasks,
         scheduler: ThreadScheduler,
         on_thread_known: Callable[[ResumeToken, anyio.Event], Awaitable[None]] | None,
+        engine_overrides_resolver: Callable[
+            [EngineId], Awaitable[EngineRunOptions | None]
+        ]
+        | None,
         chat_id: int,
         user_msg_id: int,
         thread_id: int | None,
@@ -247,6 +306,7 @@ class _TelegramCommandExecutor(CommandExecutor):
         self._running_tasks = running_tasks
         self._scheduler = scheduler
         self._on_thread_known = on_thread_known
+        self._engine_overrides_resolver = engine_overrides_resolver
         self._chat_id = chat_id
         self._user_msg_id = user_msg_id
         self._thread_id = thread_id
@@ -317,6 +377,9 @@ class _TelegramCommandExecutor(CommandExecutor):
             engine_override=request.engine,
             context=request.context,
         )
+        run_options = None
+        if self._engine_overrides_resolver is not None:
+            run_options = await self._engine_overrides_resolver(engine)
         on_thread_known = (
             self._scheduler.note_thread_known
             if self._on_thread_known is None
@@ -343,6 +406,7 @@ class _TelegramCommandExecutor(CommandExecutor):
                 engine_override=engine,
                 thread_id=self._thread_id,
                 show_resume_line=effective_show_resume_line,
+                run_options=run_options,
             )
             return RunResult(engine=engine, message=capture.last_message)
         await _run_engine(
@@ -359,6 +423,7 @@ class _TelegramCommandExecutor(CommandExecutor):
             engine_override=engine,
             thread_id=self._thread_id,
             show_resume_line=effective_show_resume_line,
+            run_options=run_options,
         )
         return RunResult(engine=engine, message=None)
 

src/takopi/telegram/commands/menu.py

@@ -73,6 +73,8 @@ def build_bot_commands(
     for cmd, description in [
         ("new", "start a new thread"),
         ("agent", "set default agent"),
+        ("model", "set model override"),
+        ("reasoning", "set reasoning override"),
         ("trigger", "set trigger mode"),
     ]:
         if cmd in seen:

src/takopi/telegram/commands/model.py

@@ -0,0 +1,284 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+from ...context import RunContext
+from ...directives import DirectiveError
+from ..chat_prefs import ChatPrefsStore
+from ..engine_defaults import resolve_engine_for_message
+from ..engine_overrides import EngineOverrides, resolve_override_value
+from ..files import split_command_args
+from ..topic_state import TopicStateStore
+from ..topics import _topic_key
+from ..types import TelegramIncomingMessage
+from .reply import make_reply
+
+if TYPE_CHECKING:
+    from ..bridge import TelegramBridgeConfig
+
+MODEL_USAGE = (
+    "usage: `/model`, `/model set <model>`, "
+    "`/model set <engine> <model>`, or `/model clear [engine]`"
+)
+
+
+async def _check_model_permissions(
+    cfg: TelegramBridgeConfig, msg: TelegramIncomingMessage
+) -> bool:
+    reply = make_reply(cfg, msg)
+    sender_id = msg.sender_id
+    if sender_id is None:
+        await reply(text="cannot verify sender for model overrides.")
+        return False
+    is_private = msg.chat_type == "private"
+    if msg.chat_type is None:
+        is_private = msg.chat_id > 0
+    if is_private:
+        return True
+    member = await cfg.bot.get_chat_member(msg.chat_id, sender_id)
+    if member is None:
+        await reply(text="failed to verify model override permissions.")
+        return False
+    if member.status in {"creator", "administrator"}:
+        return True
+    await reply(text="changing model overrides is restricted to group admins.")
+    return False
+
+
+async def _resolve_engine_selection(
+    cfg: TelegramBridgeConfig,
+    msg: TelegramIncomingMessage,
+    *,
+    ambient_context: RunContext | None,
+    topic_store: TopicStateStore | None,
+    chat_prefs: ChatPrefsStore | None,
+    topic_key: tuple[int, int] | None,
+) -> tuple[str, str] | None:
+    reply = make_reply(cfg, msg)
+    try:
+        resolved = cfg.runtime.resolve_message(
+            text="",
+            reply_text=msg.reply_to_text,
+            ambient_context=ambient_context,
+            chat_id=msg.chat_id,
+        )
+    except DirectiveError as exc:
+        await reply(text=f"error:\n{exc}")
+        return None
+    selection = await resolve_engine_for_message(
+        runtime=cfg.runtime,
+        context=resolved.context,
+        explicit_engine=None,
+        chat_id=msg.chat_id,
+        topic_key=topic_key,
+        topic_store=topic_store,
+        chat_prefs=chat_prefs,
+    )
+    return selection.engine, selection.source
+
+
+def _parse_set_args(
+    tokens: tuple[str, ...], *, engine_ids: set[str]
+) -> tuple[str | None, str | None]:
+    if len(tokens) < 2:
+        return None, None
+    if len(tokens) == 2:
+        maybe_engine = tokens[1].strip().lower()
+        if maybe_engine in engine_ids:
+            return None, None
+        return None, tokens[1].strip()
+    maybe_engine = tokens[1].strip().lower()
+    if maybe_engine in engine_ids:
+        model = " ".join(tokens[2:]).strip()
+        return maybe_engine, model or None
+    model = " ".join(tokens[1:]).strip()
+    return None, model or None
+
+
+async def _handle_model_command(
+    cfg: TelegramBridgeConfig,
+    msg: TelegramIncomingMessage,
+    args_text: str,
+    ambient_context: RunContext | None,
+    topic_store: TopicStateStore | None,
+    chat_prefs: ChatPrefsStore | None,
+    *,
+    resolved_scope: str | None = None,
+    scope_chat_ids: frozenset[int] | None = None,
+) -> None:
+    reply = make_reply(cfg, msg)
+    tkey = (
+        _topic_key(msg, cfg, scope_chat_ids=scope_chat_ids)
+        if topic_store is not None
+        else None
+    )
+    tokens = split_command_args(args_text)
+    action = tokens[0].lower() if tokens else "show"
+    engine_ids = {engine.lower() for engine in cfg.runtime.engine_ids}
+
+    if action in {"show", ""}:
+        selection = await _resolve_engine_selection(
+            cfg,
+            msg,
+            ambient_context=ambient_context,
+            topic_store=topic_store,
+            chat_prefs=chat_prefs,
+            topic_key=tkey,
+        )
+        if selection is None:
+            return
+        engine, engine_source = selection
+        topic_override = None
+        if tkey is not None and topic_store is not None:
+            topic_override = await topic_store.get_engine_override(
+                tkey[0], tkey[1], engine
+            )
+        chat_override = None
+        if chat_prefs is not None:
+            chat_override = await chat_prefs.get_engine_override(msg.chat_id, engine)
+        resolution = resolve_override_value(
+            topic_override=topic_override,
+            chat_override=chat_override,
+            field="model",
+        )
+        source_labels = {
+            "directive": "directive",
+            "topic_default": "topic default",
+            "chat_default": "chat default",
+            "project_default": "project default",
+            "global_default": "global default",
+        }
+        override_labels = {
+            "topic_override": "topic override",
+            "chat_default": "chat default",
+            "default": "no override",
+        }
+        engine_line = f"engine: {engine} ({source_labels[engine_source]})"
+        model_value = resolution.value or "default"
+        model_line = f"model: {model_value} ({override_labels[resolution.source]})"
+        topic_label = resolution.topic_value or "none"
+        if tkey is None:
+            topic_label = "none"
+        chat_label = (
+            "unavailable" if chat_prefs is None else resolution.chat_value or "none"
+        )
+        defaults_line = f"defaults: topic: {topic_label}, chat: {chat_label}"
+        available_line = f"available engines: {', '.join(cfg.runtime.engine_ids)}"
+        await reply(
+            text="\n\n".join([engine_line, model_line, defaults_line, available_line])
+        )
+        return
+
+    if action == "set":
+        engine_arg, model = _parse_set_args(tokens, engine_ids=engine_ids)
+        if model is None:
+            await reply(text=MODEL_USAGE)
+            return
+        if not await _check_model_permissions(cfg, msg):
+            return
+        if engine_arg is None:
+            selection = await _resolve_engine_selection(
+                cfg,
+                msg,
+                ambient_context=ambient_context,
+                topic_store=topic_store,
+                chat_prefs=chat_prefs,
+                topic_key=tkey,
+            )
+            if selection is None:
+                return
+            engine, _ = selection
+        else:
+            engine = engine_arg
+            if engine not in engine_ids:
+                available = ", ".join(cfg.runtime.engine_ids)
+                await reply(
+                    text=f"unknown engine `{engine}`.\navailable agents: `{available}`"
+                )
+                return
+        if tkey is not None:
+            if topic_store is None:
+                await reply(text="topic model overrides are unavailable.")
+                return
+            current = await topic_store.get_engine_override(tkey[0], tkey[1], engine)
+            updated = EngineOverrides(
+                model=model,
+                reasoning=current.reasoning if current is not None else None,
+            )
+            await topic_store.set_engine_override(tkey[0], tkey[1], engine, updated)
+            await reply(
+                text=(
+                    f"topic model override set to `{model}` for `{engine}`.\n"
+                    "If you want a clean start on the new model, run `/new`."
+                )
+            )
+            return
+        if chat_prefs is None:
+            await reply(text="chat model overrides are unavailable (no config path).")
+            return
+        current = await chat_prefs.get_engine_override(msg.chat_id, engine)
+        updated = EngineOverrides(
+            model=model,
+            reasoning=current.reasoning if current is not None else None,
+        )
+        await chat_prefs.set_engine_override(msg.chat_id, engine, updated)
+        await reply(
+            text=(
+                f"chat model override set to `{model}` for `{engine}`.\n"
+                "If you want a clean start on the new model, run `/new`."
+            )
+        )
+        return
+
+    if action == "clear":
+        engine = None
+        if len(tokens) > 2:
+            await reply(text=MODEL_USAGE)
+            return
+        if len(tokens) == 2:
+            engine = tokens[1].strip().lower() or None
+        if not await _check_model_permissions(cfg, msg):
+            return
+        if engine is None:
+            selection = await _resolve_engine_selection(
+                cfg,
+                msg,
+                ambient_context=ambient_context,
+                topic_store=topic_store,
+                chat_prefs=chat_prefs,
+                topic_key=tkey,
+            )
+            if selection is None:
+                return
+            engine, _ = selection
+        if engine not in engine_ids:
+            available = ", ".join(cfg.runtime.engine_ids)
+            await reply(
+                text=f"unknown engine `{engine}`.\navailable agents: `{available}`"
+            )
+            return
+        if tkey is not None:
+            if topic_store is None:
+                await reply(text="topic model overrides are unavailable.")
+                return
+            current = await topic_store.get_engine_override(tkey[0], tkey[1], engine)
+            updated = EngineOverrides(
+                model=None,
+                reasoning=current.reasoning if current is not None else None,
+            )
+            await topic_store.set_engine_override(tkey[0], tkey[1], engine, updated)
+            await reply(text="topic model override cleared (using chat default).")
+            return
+        if chat_prefs is None:
+            await reply(text="chat model overrides are unavailable (no config path).")
+            return
+        current = await chat_prefs.get_engine_override(msg.chat_id, engine)
+        updated = EngineOverrides(
+            model=None,
+            reasoning=current.reasoning if current is not None else None,
+        )
+        await chat_prefs.set_engine_override(msg.chat_id, engine, updated)
+        await reply(text="chat model override cleared.")
+        return
+
+    await reply(text=MODEL_USAGE)

src/takopi/telegram/commands/reasoning.py

@@ -0,0 +1,308 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+from ...context import RunContext
+from ...directives import DirectiveError
+from ..chat_prefs import ChatPrefsStore
+from ..engine_defaults import resolve_engine_for_message
+from ..engine_overrides import (
+    EngineOverrides,
+    allowed_reasoning_levels,
+    resolve_override_value,
+)
+from ..files import split_command_args
+from ..topic_state import TopicStateStore
+from ..topics import _topic_key
+from ..types import TelegramIncomingMessage
+from .reply import make_reply
+
+if TYPE_CHECKING:
+    from ..bridge import TelegramBridgeConfig
+
+REASONING_USAGE = (
+    "usage: `/reasoning`, `/reasoning set <level>`, "
+    "`/reasoning set <engine> <level>`, or `/reasoning clear [engine]`"
+)
+
+
+async def _check_reasoning_permissions(
+    cfg: TelegramBridgeConfig, msg: TelegramIncomingMessage
+) -> bool:
+    reply = make_reply(cfg, msg)
+    sender_id = msg.sender_id
+    if sender_id is None:
+        await reply(text="cannot verify sender for reasoning overrides.")
+        return False
+    is_private = msg.chat_type == "private"
+    if msg.chat_type is None:
+        is_private = msg.chat_id > 0
+    if is_private:
+        return True
+    member = await cfg.bot.get_chat_member(msg.chat_id, sender_id)
+    if member is None:
+        await reply(text="failed to verify reasoning override permissions.")
+        return False
+    if member.status in {"creator", "administrator"}:
+        return True
+    await reply(text="changing reasoning overrides is restricted to group admins.")
+    return False
+
+
+async def _resolve_engine_selection(
+    cfg: TelegramBridgeConfig,
+    msg: TelegramIncomingMessage,
+    *,
+    ambient_context: RunContext | None,
+    topic_store: TopicStateStore | None,
+    chat_prefs: ChatPrefsStore | None,
+    topic_key: tuple[int, int] | None,
+) -> tuple[str, str] | None:
+    reply = make_reply(cfg, msg)
+    try:
+        resolved = cfg.runtime.resolve_message(
+            text="",
+            reply_text=msg.reply_to_text,
+            ambient_context=ambient_context,
+            chat_id=msg.chat_id,
+        )
+    except DirectiveError as exc:
+        await reply(text=f"error:\n{exc}")
+        return None
+    selection = await resolve_engine_for_message(
+        runtime=cfg.runtime,
+        context=resolved.context,
+        explicit_engine=None,
+        chat_id=msg.chat_id,
+        topic_key=topic_key,
+        topic_store=topic_store,
+        chat_prefs=chat_prefs,
+    )
+    return selection.engine, selection.source
+
+
+def _parse_set_args(
+    tokens: tuple[str, ...], *, engine_ids: set[str]
+) -> tuple[str | None, str | None]:
+    if len(tokens) < 2:
+        return None, None
+    if len(tokens) == 2:
+        maybe_engine = tokens[1].strip().lower()
+        if maybe_engine in engine_ids:
+            return None, None
+        return None, tokens[1].strip()
+    maybe_engine = tokens[1].strip().lower()
+    if maybe_engine in engine_ids:
+        level = " ".join(tokens[2:]).strip()
+        return maybe_engine, level or None
+    level = " ".join(tokens[1:]).strip()
+    return None, level or None
+
+
+async def _handle_reasoning_command(
+    cfg: TelegramBridgeConfig,
+    msg: TelegramIncomingMessage,
+    args_text: str,
+    ambient_context: RunContext | None,
+    topic_store: TopicStateStore | None,
+    chat_prefs: ChatPrefsStore | None,
+    *,
+    resolved_scope: str | None = None,
+    scope_chat_ids: frozenset[int] | None = None,
+) -> None:
+    reply = make_reply(cfg, msg)
+    tkey = (
+        _topic_key(msg, cfg, scope_chat_ids=scope_chat_ids)
+        if topic_store is not None
+        else None
+    )
+    tokens = split_command_args(args_text)
+    action = tokens[0].lower() if tokens else "show"
+    engine_ids = {engine.lower() for engine in cfg.runtime.engine_ids}
+
+    if action in {"show", ""}:
+        selection = await _resolve_engine_selection(
+            cfg,
+            msg,
+            ambient_context=ambient_context,
+            topic_store=topic_store,
+            chat_prefs=chat_prefs,
+            topic_key=tkey,
+        )
+        if selection is None:
+            return
+        engine, engine_source = selection
+        topic_override = None
+        if tkey is not None and topic_store is not None:
+            topic_override = await topic_store.get_engine_override(
+                tkey[0], tkey[1], engine
+            )
+        chat_override = None
+        if chat_prefs is not None:
+            chat_override = await chat_prefs.get_engine_override(msg.chat_id, engine)
+        resolution = resolve_override_value(
+            topic_override=topic_override,
+            chat_override=chat_override,
+            field="reasoning",
+        )
+        source_labels = {
+            "directive": "directive",
+            "topic_default": "topic default",
+            "chat_default": "chat default",
+            "project_default": "project default",
+            "global_default": "global default",
+        }
+        override_labels = {
+            "topic_override": "topic override",
+            "chat_default": "chat default",
+            "default": "no override",
+        }
+        engine_line = f"engine: {engine} ({source_labels[engine_source]})"
+        reasoning_value = resolution.value or "default"
+        reasoning_line = (
+            f"reasoning: {reasoning_value} ({override_labels[resolution.source]})"
+        )
+        topic_label = resolution.topic_value or "none"
+        if tkey is None:
+            topic_label = "none"
+        chat_label = (
+            "unavailable" if chat_prefs is None else resolution.chat_value or "none"
+        )
+        defaults_line = f"defaults: topic: {topic_label}, chat: {chat_label}"
+        available_levels = ", ".join(allowed_reasoning_levels(engine))
+        available_line = f"available levels: {available_levels}"
+        await reply(
+            text="\n\n".join(
+                [engine_line, reasoning_line, defaults_line, available_line]
+            )
+        )
+        return
+
+    if action == "set":
+        engine_arg, level = _parse_set_args(tokens, engine_ids=engine_ids)
+        if level is None:
+            await reply(text=REASONING_USAGE)
+            return
+        if not await _check_reasoning_permissions(cfg, msg):
+            return
+        if engine_arg is None:
+            selection = await _resolve_engine_selection(
+                cfg,
+                msg,
+                ambient_context=ambient_context,
+                topic_store=topic_store,
+                chat_prefs=chat_prefs,
+                topic_key=tkey,
+            )
+            if selection is None:
+                return
+            engine, _ = selection
+        else:
+            engine = engine_arg
+            if engine not in engine_ids:
+                available = ", ".join(cfg.runtime.engine_ids)
+                await reply(
+                    text=f"unknown engine `{engine}`.\navailable agents: `{available}`"
+                )
+                return
+        normalized_level = level.strip().lower()
+        allowed = allowed_reasoning_levels(engine)
+        if normalized_level not in allowed:
+            await reply(
+                text=(
+                    f"unknown reasoning level `{level}`.\n"
+                    f"available levels: {', '.join(allowed)}"
+                )
+            )
+            return
+        if tkey is not None:
+            if topic_store is None:
+                await reply(text="topic reasoning overrides are unavailable.")
+                return
+            current = await topic_store.get_engine_override(tkey[0], tkey[1], engine)
+            updated = EngineOverrides(
+                model=current.model if current is not None else None,
+                reasoning=normalized_level,
+            )
+            await topic_store.set_engine_override(tkey[0], tkey[1], engine, updated)
+            await reply(
+                text=(
+                    f"topic reasoning override set to `{normalized_level}` "
+                    f"for `{engine}`.\n"
+                    "If you want a clean start on the new setting, run `/new`."
+                )
+            )
+            return
+        if chat_prefs is None:
+            await reply(
+                text="chat reasoning overrides are unavailable (no config path)."
+            )
+            return
+        current = await chat_prefs.get_engine_override(msg.chat_id, engine)
+        updated = EngineOverrides(
+            model=current.model if current is not None else None,
+            reasoning=normalized_level,
+        )
+        await chat_prefs.set_engine_override(msg.chat_id, engine, updated)
+        await reply(
+            text=(
+                f"chat reasoning override set to `{normalized_level}` for `{engine}`.\n"
+                "If you want a clean start on the new setting, run `/new`."
+            )
+        )
+        return
+
+    if action == "clear":
+        engine = None
+        if len(tokens) > 2:
+            await reply(text=REASONING_USAGE)
+            return
+        if len(tokens) == 2:
+            engine = tokens[1].strip().lower() or None
+        if not await _check_reasoning_permissions(cfg, msg):
+            return
+        if engine is None:
+            selection = await _resolve_engine_selection(
+                cfg,
+                msg,
+                ambient_context=ambient_context,
+                topic_store=topic_store,
+                chat_prefs=chat_prefs,
+                topic_key=tkey,
+            )
+            if selection is None:
+                return
+            engine, _ = selection
+        if engine not in engine_ids:
+            available = ", ".join(cfg.runtime.engine_ids)
+            await reply(
+                text=f"unknown engine `{engine}`.\navailable agents: `{available}`"
+            )
+            return
+        if tkey is not None:
+            if topic_store is None:
+                await reply(text="topic reasoning overrides are unavailable.")
+                return
+            current = await topic_store.get_engine_override(tkey[0], tkey[1], engine)
+            updated = EngineOverrides(
+                model=current.model if current is not None else None,
+                reasoning=None,
+            )
+            await topic_store.set_engine_override(tkey[0], tkey[1], engine, updated)
+            await reply(text="topic reasoning override cleared (using chat default).")
+            return
+        if chat_prefs is None:
+            await reply(
+                text="chat reasoning overrides are unavailable (no config path)."
+            )
+            return
+        current = await chat_prefs.get_engine_override(msg.chat_id, engine)
+        updated = EngineOverrides(
+            model=current.model if current is not None else None,
+            reasoning=None,
+        )
+        await chat_prefs.set_engine_override(msg.chat_id, engine, updated)
+        await reply(text="chat reasoning override cleared.")
+        return
+
+    await reply(text=REASONING_USAGE)

src/takopi/telegram/engine_overrides.py

@@ -0,0 +1,107 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Literal
+
+import msgspec
+
+OverrideSource = Literal["topic_override", "chat_default", "default"]
+
+REASONING_LEVELS: tuple[str, ...] = ("minimal", "low", "medium", "high", "xhigh")
+OPENCODE_REASONING_LEVELS: tuple[str, ...] = ("none", *REASONING_LEVELS)
+REASONING_SUPPORTED_ENGINES = frozenset({"codex"})
+
+
+class EngineOverrides(msgspec.Struct, forbid_unknown_fields=False):
+    model: str | None = None
+    reasoning: str | None = None
+
+
+@dataclass(frozen=True, slots=True)
+class OverrideValueResolution:
+    value: str | None
+    source: OverrideSource
+    topic_value: str | None
+    chat_value: str | None
+
+
+def normalize_override_value(value: str | None) -> str | None:
+    if value is None:
+        return None
+    cleaned = value.strip()
+    return cleaned or None
+
+
+def normalize_overrides(overrides: EngineOverrides | None) -> EngineOverrides | None:
+    if overrides is None:
+        return None
+    model = normalize_override_value(overrides.model)
+    reasoning = normalize_override_value(overrides.reasoning)
+    if model is None and reasoning is None:
+        return None
+    return EngineOverrides(model=model, reasoning=reasoning)
+
+
+def merge_overrides(
+    topic_override: EngineOverrides | None,
+    chat_override: EngineOverrides | None,
+) -> EngineOverrides | None:
+    topic = normalize_overrides(topic_override)
+    chat = normalize_overrides(chat_override)
+    if topic is None and chat is None:
+        return None
+    model = None
+    reasoning = None
+    if topic is not None and topic.model is not None:
+        model = topic.model
+    elif chat is not None:
+        model = chat.model
+    if topic is not None and topic.reasoning is not None:
+        reasoning = topic.reasoning
+    elif chat is not None:
+        reasoning = chat.reasoning
+    return normalize_overrides(EngineOverrides(model=model, reasoning=reasoning))
+
+
+def resolve_override_value(
+    *,
+    topic_override: EngineOverrides | None,
+    chat_override: EngineOverrides | None,
+    field: Literal["model", "reasoning"],
+) -> OverrideValueResolution:
+    topic_value = normalize_override_value(
+        getattr(topic_override, field, None) if topic_override is not None else None
+    )
+    chat_value = normalize_override_value(
+        getattr(chat_override, field, None) if chat_override is not None else None
+    )
+    if topic_value is not None:
+        return OverrideValueResolution(
+            value=topic_value,
+            source="topic_override",
+            topic_value=topic_value,
+            chat_value=chat_value,
+        )
+    if chat_value is not None:
+        return OverrideValueResolution(
+            value=chat_value,
+            source="chat_default",
+            topic_value=topic_value,
+            chat_value=chat_value,
+        )
+    return OverrideValueResolution(
+        value=None,
+        source="default",
+        topic_value=topic_value,
+        chat_value=chat_value,
+    )
+
+
+def allowed_reasoning_levels(engine: str) -> tuple[str, ...]:
+    if engine == "opencode":
+        return OPENCODE_REASONING_LEVELS
+    return REASONING_LEVELS
+
+
+def supports_reasoning(engine: str) -> bool:
+    return engine in REASONING_SUPPORTED_ENGINES

src/takopi/telegram/loop.py

@@ -14,6 +14,7 @@ from ..commands import list_command_ids
 from ..directives import DirectiveError
 from ..logging import get_logger
 from ..model import EngineId, ResumeToken
+from ..runners.run_options import EngineRunOptions
 from ..scheduler import ThreadJob, ThreadScheduler
 from ..progress import ProgressTracker
 from ..settings import TelegramTransportSettings
@@ -42,6 +43,8 @@ from .commands.topics import (
     _handle_new_command,
     _handle_topic_command,
 )
+from .commands.model import _handle_model_command
+from .commands.reasoning import _handle_reasoning_command
 from .commands.trigger import _handle_trigger_command
 from .context import _merge_topic_context, _usage_ctx_set, _usage_topic
 from .topics import (
@@ -55,6 +58,7 @@ from .topics import (
 from .client import poll_incoming
 from .chat_prefs import ChatPrefsStore, resolve_prefs_path
 from .chat_sessions import ChatSessionStore, resolve_sessions_path
+from .engine_overrides import merge_overrides
 from .engine_defaults import resolve_engine_for_message
 from .topic_state import TopicStateStore, resolve_state_path
 from .trigger_mode import resolve_trigger_mode, should_trigger_run
@@ -86,6 +90,27 @@ def _chat_session_key(
     return (msg.chat_id, msg.sender_id)
 
 
+async def _resolve_engine_run_options(
+    chat_id: int,
+    thread_id: int | None,
+    engine: EngineId,
+    chat_prefs: ChatPrefsStore | None,
+    topic_store: TopicStateStore | None,
+) -> EngineRunOptions | None:
+    topic_override = None
+    if topic_store is not None and thread_id is not None:
+        topic_override = await topic_store.get_engine_override(
+            chat_id, thread_id, engine
+        )
+    chat_override = None
+    if chat_prefs is not None:
+        chat_override = await chat_prefs.get_engine_override(chat_id, engine)
+    merged = merge_overrides(topic_override, chat_override)
+    if merged is None:
+        return None
+    return EngineRunOptions(model=merged.model, reasoning=merged.reasoning)
+
+
 def _allowed_chat_ids(cfg: TelegramBridgeConfig) -> set[int]:
     allowed = set(cfg.chat_ids or ())
     allowed.add(cfg.chat_id)
@@ -187,6 +212,32 @@ def _dispatch_builtin_command(
             scope_chat_ids=scope_chat_ids,
         )
 
+    if command_id == "model":
+        handlers["model"] = partial(
+            _handle_model_command,
+            cfg,
+            msg,
+            args_text,
+            ambient_context,
+            topic_store,
+            chat_prefs,
+            resolved_scope=resolved_scope,
+            scope_chat_ids=scope_chat_ids,
+        )
+
+    if command_id == "reasoning":
+        handlers["reasoning"] = partial(
+            _handle_reasoning_command,
+            cfg,
+            msg,
+            args_text,
+            ambient_context,
+            topic_store,
+            chat_prefs,
+            resolved_scope=resolved_scope,
+            scope_chat_ids=scope_chat_ids,
+        )
+
     if command_id == "trigger":
         handlers["trigger"] = partial(
             _handle_trigger_command,
@@ -605,6 +656,24 @@ async def run_main_loop(
                     stateful_mode=stateful_mode,
                     context=context,
                 )
+                engine_for_overrides = (
+                    resume_token.engine
+                    if resume_token is not None
+                    else engine_override
+                    if engine_override is not None
+                    else cfg.runtime.resolve_engine(
+                        engine_override=None,
+                        context=context,
+                    )
+                )
+                overrides_thread_id = topic_key[1] if topic_key is not None else None
+                run_options = await _resolve_engine_run_options(
+                    chat_id,
+                    overrides_thread_id,
+                    engine_for_overrides,
+                    chat_prefs=chat_prefs,
+                    topic_store=topic_store,
+                )
                 await _run_engine(
                     exec_cfg=cfg.exec_cfg,
                     runtime=cfg.runtime,
@@ -622,6 +691,7 @@ async def run_main_loop(
                     thread_id=thread_id,
                     show_resume_line=show_resume_line,
                     progress_ref=progress_ref,
+                    run_options=run_options,
                 )
 
             async def run_thread_job(job: ThreadJob) -> None:
@@ -1377,6 +1447,16 @@ async def run_main_loop(
                             in {"directive", "topic_default", "chat_default"}
                             else None
                         )
+                        overrides_thread_id = (
+                            topic_key[1] if topic_key is not None else None
+                        )
+                        engine_overrides_resolver = partial(
+                            _resolve_engine_run_options,
+                            chat_id,
+                            overrides_thread_id,
+                            chat_prefs=chat_prefs,
+                            topic_store=topic_store,
+                        )
                         tg.start_soon(
                             _dispatch_command,
                             cfg,
@@ -1393,6 +1473,7 @@ async def run_main_loop(
                             ),
                             stateful_mode,
                             default_engine_override,
+                            engine_overrides_resolver,
                         )
                         continue
 

src/takopi/telegram/topic_state.py

@@ -8,6 +8,7 @@ import msgspec
 from ..context import RunContext
 from ..logging import get_logger
 from ..model import ResumeToken
+from .engine_overrides import EngineOverrides, normalize_overrides
 from .state_store import JsonStateStore
 
 logger = get_logger(__name__)
@@ -41,6 +42,7 @@ class _ThreadState(msgspec.Struct, forbid_unknown_fields=False):
     topic_title: str | None = None
     default_engine: str | None = None
     trigger_mode: str | None = None
+    engine_overrides: dict[str, EngineOverrides] = msgspec.field(default_factory=dict)
 
 
 class _TopicState(msgspec.Struct, forbid_unknown_fields=False):
@@ -74,6 +76,13 @@ def _normalize_trigger_mode(value: str | None) -> str | None:
     return None
 
 
+def _normalize_engine_id(value: str | None) -> str | None:
+    if value is None:
+        return None
+    value = value.strip().lower()
+    return value or None
+
+
 def _context_from_state(state: _ContextState | None) -> RunContext | None:
     if state is None:
         return None
@@ -181,6 +190,20 @@ class TopicStateStore(JsonStateStore[_TopicState]):
                 return None
             return _normalize_trigger_mode(thread.trigger_mode)
 
+    async def get_engine_override(
+        self, chat_id: int, thread_id: int, engine: str
+    ) -> EngineOverrides | None:
+        engine_key = _normalize_engine_id(engine)
+        if engine_key is None:
+            return None
+        async with self._lock:
+            self._reload_locked_if_needed()
+            thread = self._get_thread_locked(chat_id, thread_id)
+            if thread is None:
+                return None
+            override = thread.engine_overrides.get(engine_key)
+            return normalize_overrides(override)
+
     async def set_default_engine(
         self, chat_id: int, thread_id: int, engine: str | None
     ) -> None:
@@ -207,6 +230,31 @@ class TopicStateStore(JsonStateStore[_TopicState]):
     async def clear_trigger_mode(self, chat_id: int, thread_id: int) -> None:
         await self.set_trigger_mode(chat_id, thread_id, None)
 
+    async def set_engine_override(
+        self,
+        chat_id: int,
+        thread_id: int,
+        engine: str,
+        override: EngineOverrides | None,
+    ) -> None:
+        engine_key = _normalize_engine_id(engine)
+        if engine_key is None:
+            return
+        normalized = normalize_overrides(override)
+        async with self._lock:
+            self._reload_locked_if_needed()
+            thread = self._ensure_thread_locked(chat_id, thread_id)
+            if normalized is None:
+                thread.engine_overrides.pop(engine_key, None)
+            else:
+                thread.engine_overrides[engine_key] = normalized
+            self._save_locked()
+
+    async def clear_engine_override(
+        self, chat_id: int, thread_id: int, engine: str
+    ) -> None:
+        await self.set_engine_override(chat_id, thread_id, engine, None)
+
     async def set_session_resume(
         self, chat_id: int, thread_id: int, token: ResumeToken
     ) -> None:

tests/test_runner_run_options.py

@@ -0,0 +1,59 @@
+from takopi.model import ResumeToken
+from takopi.runners.claude import ClaudeRunner
+from takopi.runners.codex import CodexRunner
+from takopi.runners.opencode import OpenCodeRunner, OpenCodeStreamState
+from takopi.runners.pi import ENGINE as PI_ENGINE, PiRunner, PiStreamState
+from takopi.runners.run_options import EngineRunOptions, apply_run_options
+
+
+def test_codex_run_options_override_model_and_reasoning() -> None:
+    runner = CodexRunner(codex_cmd="codex", extra_args=["-c", "notify=[]"])
+    state = runner.new_state("hi", None)
+    with apply_run_options(EngineRunOptions(model="gpt-4.1-mini", reasoning="low")):
+        args = runner.build_args("hi", None, state=state)
+
+    assert args == [
+        "-c",
+        "notify=[]",
+        "--model",
+        "gpt-4.1-mini",
+        "-c",
+        "model_reasoning_effort=low",
+        "exec",
+        "--json",
+        "--skip-git-repo-check",
+        "--color=never",
+        "-",
+    ]
+
+
+def test_claude_run_options_override_model() -> None:
+    runner = ClaudeRunner(claude_cmd="claude", model="claude-sonnet")
+    with apply_run_options(EngineRunOptions(model="claude-opus")):
+        args = runner.build_args("hi", None, state=None)
+
+    assert "--model" in args
+    model_idx = args.index("--model") + 1
+    assert args[model_idx] == "claude-opus"
+
+
+def test_opencode_run_options_override_model() -> None:
+    runner = OpenCodeRunner(opencode_cmd="opencode", model="claude-sonnet")
+    state = OpenCodeStreamState()
+    with apply_run_options(EngineRunOptions(model="gpt-4o-mini")):
+        args = runner.build_args("hi", None, state=state)
+
+    assert "--model" in args
+    model_idx = args.index("--model") + 1
+    assert args[model_idx] == "gpt-4o-mini"
+
+
+def test_pi_run_options_override_model() -> None:
+    runner = PiRunner(extra_args=[], model="pi-default", provider=None)
+    state = PiStreamState(resume=ResumeToken(engine=PI_ENGINE, value="sess.jsonl"))
+    with apply_run_options(EngineRunOptions(model="pi-override")):
+        args = runner.build_args("hi", None, state=state)
+
+    assert "--model" in args
+    model_idx = args.index("--model") + 1
+    assert args[model_idx] == "pi-override"

tests/test_telegram_bridge.py

@@ -8,6 +8,8 @@ import pytest
 from takopi import commands, plugins
 from takopi.telegram.commands.executor import _CaptureTransport, _run_engine
 from takopi.telegram.commands.file_transfer import _handle_file_get, _handle_file_put
+from takopi.telegram.commands.model import _handle_model_command
+from takopi.telegram.commands.reasoning import _handle_reasoning_command
 from takopi.telegram.commands.topics import _handle_topic_command
 import takopi.telegram.loop as telegram_loop
 import takopi.telegram.topics as telegram_topics
@@ -38,6 +40,7 @@ from takopi.telegram.render import MAX_BODY_CHARS
 from takopi.telegram.topic_state import TopicStateStore, resolve_state_path
 from takopi.telegram.chat_sessions import ChatSessionStore, resolve_sessions_path
 from takopi.telegram.chat_prefs import ChatPrefsStore, resolve_prefs_path
+from takopi.telegram.engine_overrides import EngineOverrides
 from takopi.context import RunContext
 from takopi.config import ProjectConfig, ProjectsConfig
 from takopi.runner_bridge import ExecBridgeConfig, RunningTask
@@ -1348,6 +1351,230 @@ async def test_topic_command_recreates_stale_topic(tmp_path: Path) -> None:
     assert snapshot.context == RunContext(project="takopi", branch="master")
 
 
+@pytest.mark.anyio
+async def test_model_command_show_reports_overrides(tmp_path: Path) -> None:
+    transport = _FakeTransport()
+    cfg = _make_cfg(transport)
+    cfg = replace(cfg, topics=TelegramTopicsSettings(enabled=True, scope="main"))
+    chat_prefs = ChatPrefsStore(tmp_path / "telegram_chat_prefs_state.json")
+    topic_store = TopicStateStore(tmp_path / "telegram_topics_state.json")
+    await chat_prefs.set_engine_override(
+        123,
+        CODEX_ENGINE,
+        EngineOverrides(model="gpt-4.1-mini", reasoning=None),
+    )
+    await topic_store.set_engine_override(
+        123,
+        77,
+        CODEX_ENGINE,
+        EngineOverrides(model="gpt-4.1", reasoning=None),
+    )
+    msg = TelegramIncomingMessage(
+        transport="telegram",
+        chat_id=123,
+        message_id=10,
+        text="/model",
+        reply_to_message_id=None,
+        reply_to_text=None,
+        sender_id=123,
+        thread_id=77,
+    )
+
+    await _handle_model_command(
+        cfg,
+        msg,
+        "",
+        ambient_context=None,
+        topic_store=topic_store,
+        chat_prefs=chat_prefs,
+        resolved_scope="main",
+        scope_chat_ids=frozenset({123}),
+    )
+
+    text = transport.send_calls[-1]["message"].text
+    assert "engine: codex (global default)" in text
+    assert "model: gpt-4.1 (topic override)" in text
+    assert "defaults: topic: gpt-4.1, chat: gpt-4.1-mini" in text
+    assert "available engines: codex" in text
+
+
+@pytest.mark.anyio
+async def test_model_command_set_and_clear_chat_override(tmp_path: Path) -> None:
+    transport = _FakeTransport()
+    cfg = _make_cfg(transport)
+    chat_prefs = ChatPrefsStore(tmp_path / "telegram_chat_prefs_state.json")
+    await chat_prefs.set_engine_override(
+        123,
+        CODEX_ENGINE,
+        EngineOverrides(model=None, reasoning="low"),
+    )
+    msg = TelegramIncomingMessage(
+        transport="telegram",
+        chat_id=123,
+        message_id=10,
+        text="/model set gpt-4.1-mini",
+        reply_to_message_id=None,
+        reply_to_text=None,
+        sender_id=456,
+        chat_type="supergroup",
+    )
+
+    await _handle_model_command(
+        cfg,
+        msg,
+        "set gpt-4.1-mini",
+        ambient_context=None,
+        topic_store=None,
+        chat_prefs=chat_prefs,
+    )
+
+    override = await chat_prefs.get_engine_override(123, CODEX_ENGINE)
+    assert override is not None
+    assert override.model == "gpt-4.1-mini"
+    assert override.reasoning == "low"
+    assert (
+        "chat model override set to gpt-4.1-mini for codex."
+        in transport.send_calls[-1]["message"].text
+    )
+
+    msg_clear = replace(
+        msg,
+        message_id=11,
+        text="/model clear codex",
+    )
+    await _handle_model_command(
+        cfg,
+        msg_clear,
+        "clear codex",
+        ambient_context=None,
+        topic_store=None,
+        chat_prefs=chat_prefs,
+    )
+
+    override = await chat_prefs.get_engine_override(123, CODEX_ENGINE)
+    assert override is not None
+    assert override.model is None
+    assert override.reasoning == "low"
+    assert "chat model override cleared." in transport.send_calls[-1]["message"].text
+
+
+@pytest.mark.anyio
+async def test_reasoning_command_set_and_clear_topic_override(tmp_path: Path) -> None:
+    transport = _FakeTransport()
+    cfg = _make_cfg(transport)
+    cfg = replace(cfg, topics=TelegramTopicsSettings(enabled=True, scope="main"))
+    topic_store = TopicStateStore(tmp_path / "telegram_topics_state.json")
+    await topic_store.set_engine_override(
+        123,
+        77,
+        CODEX_ENGINE,
+        EngineOverrides(model="gpt-4.1-mini", reasoning=None),
+    )
+    msg = TelegramIncomingMessage(
+        transport="telegram",
+        chat_id=123,
+        message_id=10,
+        text="/reasoning set High",
+        reply_to_message_id=None,
+        reply_to_text=None,
+        sender_id=456,
+        chat_type="supergroup",
+        thread_id=77,
+    )
+
+    await _handle_reasoning_command(
+        cfg,
+        msg,
+        "set High",
+        ambient_context=None,
+        topic_store=topic_store,
+        chat_prefs=None,
+        resolved_scope="main",
+        scope_chat_ids=frozenset({123}),
+    )
+
+    override = await topic_store.get_engine_override(123, 77, CODEX_ENGINE)
+    assert override is not None
+    assert override.model == "gpt-4.1-mini"
+    assert override.reasoning == "high"
+    assert (
+        "topic reasoning override set to high for codex."
+        in transport.send_calls[-1]["message"].text
+    )
+
+    msg_clear = replace(
+        msg,
+        message_id=11,
+        text="/reasoning clear",
+    )
+    await _handle_reasoning_command(
+        cfg,
+        msg_clear,
+        "clear",
+        ambient_context=None,
+        topic_store=topic_store,
+        chat_prefs=None,
+        resolved_scope="main",
+        scope_chat_ids=frozenset({123}),
+    )
+
+    override = await topic_store.get_engine_override(123, 77, CODEX_ENGINE)
+    assert override is not None
+    assert override.model == "gpt-4.1-mini"
+    assert override.reasoning is None
+    assert (
+        "topic reasoning override cleared (using chat default)."
+        in transport.send_calls[-1]["message"].text
+    )
+
+
+@pytest.mark.anyio
+async def test_reasoning_command_show_reports_overrides(tmp_path: Path) -> None:
+    transport = _FakeTransport()
+    cfg = _make_cfg(transport)
+    cfg = replace(cfg, topics=TelegramTopicsSettings(enabled=True, scope="main"))
+    chat_prefs = ChatPrefsStore(tmp_path / "telegram_chat_prefs_state.json")
+    topic_store = TopicStateStore(tmp_path / "telegram_topics_state.json")
+    await chat_prefs.set_engine_override(
+        123,
+        CODEX_ENGINE,
+        EngineOverrides(model=None, reasoning="low"),
+    )
+    await topic_store.set_engine_override(
+        123,
+        88,
+        CODEX_ENGINE,
+        EngineOverrides(model=None, reasoning="high"),
+    )
+    msg = TelegramIncomingMessage(
+        transport="telegram",
+        chat_id=123,
+        message_id=10,
+        text="/reasoning",
+        reply_to_message_id=None,
+        reply_to_text=None,
+        sender_id=123,
+        thread_id=88,
+    )
+
+    await _handle_reasoning_command(
+        cfg,
+        msg,
+        "",
+        ambient_context=None,
+        topic_store=topic_store,
+        chat_prefs=chat_prefs,
+        resolved_scope="main",
+        scope_chat_ids=frozenset({123}),
+    )
+
+    text = transport.send_calls[-1]["message"].text
+    assert "engine: codex (global default)" in text
+    assert "reasoning: high (topic override)" in text
+    assert "defaults: topic: high, chat: low" in text
+    assert "available levels: minimal, low, medium, high, xhigh" in text
+
+
 @pytest.mark.anyio
 async def test_send_with_resume_waits_for_token() -> None:
     transport = _FakeTransport()

tests/test_telegram_engine_overrides.py

@@ -0,0 +1,97 @@
+import pytest
+
+from takopi.telegram.chat_prefs import ChatPrefsStore
+from takopi.telegram.engine_overrides import (
+    EngineOverrides,
+    merge_overrides,
+    resolve_override_value,
+)
+from takopi.telegram.topic_state import TopicStateStore
+
+
+def test_merge_overrides_prefers_topic_values() -> None:
+    topic = EngineOverrides(model=None, reasoning="high")
+    chat = EngineOverrides(model="gpt-4.1-mini", reasoning=None)
+    merged = merge_overrides(topic, chat)
+
+    assert merged is not None
+    assert merged.model == "gpt-4.1-mini"
+    assert merged.reasoning == "high"
+
+
+def test_resolve_override_value_tracks_sources() -> None:
+    topic = EngineOverrides(model="gpt-4.1", reasoning=None)
+    chat = EngineOverrides(model="gpt-4.1-mini", reasoning="low")
+    resolution = resolve_override_value(
+        topic_override=topic,
+        chat_override=chat,
+        field="model",
+    )
+
+    assert resolution.value == "gpt-4.1"
+    assert resolution.source == "topic_override"
+    assert resolution.topic_value == "gpt-4.1"
+    assert resolution.chat_value == "gpt-4.1-mini"
+
+
+@pytest.mark.anyio
+async def test_chat_prefs_engine_overrides_roundtrip(tmp_path) -> None:
+    path = tmp_path / "telegram_chat_prefs_state.json"
+    store = ChatPrefsStore(path)
+    await store.set_engine_override(
+        123,
+        "codex",
+        EngineOverrides(model="gpt-4.1-mini", reasoning="low"),
+    )
+
+    override = await store.get_engine_override(123, "codex")
+    assert override is not None
+    assert override.model == "gpt-4.1-mini"
+    assert override.reasoning == "low"
+
+    store2 = ChatPrefsStore(path)
+    override2 = await store2.get_engine_override(123, "codex")
+    assert override2 is not None
+    assert override2.model == "gpt-4.1-mini"
+    assert override2.reasoning == "low"
+
+    await store2.set_engine_override(
+        123,
+        "codex",
+        EngineOverrides(model=None, reasoning="low"),
+    )
+    override3 = await store2.get_engine_override(123, "codex")
+    assert override3 is not None
+    assert override3.model is None
+    assert override3.reasoning == "low"
+
+    await store2.set_engine_override(
+        123,
+        "codex",
+        EngineOverrides(model=None, reasoning=None),
+    )
+    override4 = await store2.get_engine_override(123, "codex")
+    assert override4 is None
+
+
+@pytest.mark.anyio
+async def test_topic_state_engine_overrides_roundtrip(tmp_path) -> None:
+    path = tmp_path / "telegram_topics_state.json"
+    store = TopicStateStore(path)
+    await store.set_engine_override(
+        1,
+        10,
+        "codex",
+        EngineOverrides(model="gpt-4.1", reasoning="medium"),
+    )
+
+    override = await store.get_engine_override(1, 10, "codex")
+    assert override is not None
+    assert override.model == "gpt-4.1"
+    assert override.reasoning == "medium"
+
+    store2 = TopicStateStore(path)
+    override2 = await store2.get_engine_override(1, 10, "codex")
+    assert override2 is not None
+    assert override2.model == "gpt-4.1"
+    assert override2.reasoning == "medium"