LLeMbas/src/lembas/api/chats.py

"""Chat creation, messaging and the streaming reply endpoint."""

from __future__ import annotations

import asyncio
import json
import logging
import time
from collections.abc import AsyncIterator
from datetime import UTC, datetime

from fastapi import APIRouter, Depends, Form, HTTPException, Request, status
from fastapi.responses import HTMLResponse, JSONResponse, Response, StreamingResponse
from sqlalchemy import func, select
from sqlalchemy.orm import Session as DBSession

from lembas.api.deps import Db, RequiredUser, require_permission
from lembas.db.models import (
    KIND_AGENT,
    KIND_CHAT,
    ROLE_ASSISTANT,
    ROLE_USER,
    Chat,
    Message,
    Model,
    User,
)
from lembas.db.session import session_scope
from lembas.security import permissions
from lembas.services import audio as audio_service
from lembas.services import chat as chat_service
from lembas.services import compaction as compaction_service
from lembas.services import files as files_service
from lembas.services import generation as generation_service
from lembas.services import metrics as metrics_service
from lembas.services import prompts as prompts_service
from lembas.services import settings_store, sse
from lembas.services import tools as tools_service
from lembas.services.agent import policy as agent_policy
from lembas.services.agent import terminal as terminal_service
from lembas.services.markdown import escape_text, render_markdown
from lembas.web.templating import render, templates

log = logging.getLogger(__name__)

router = APIRouter(prefix="/api/chats", tags=["chats"])

# Seconds of silence before a comment frame is sent to hold the connection open.
# Well under nginx's 60s default; see services/sse.py:KEEPALIVE.
KEEPALIVE_AFTER = 15.0

# How many prompts may wait behind a reply at once. The terminal panel's Auto
# send is what this exists for: a `for` loop in a shell can produce commands
# faster than any model answers them, and a bound with a sentence attached is
# better than four hundred rows nobody meant to write.
MAX_QUEUED = 10

# How many things one chat may have switched off. There are a dozen families and
# sixty skills at most, so this is not a limit anybody reaches by hand -- it is
# there so a crafted POST cannot grow the column without bound.
MAX_SCOPE_KEYS = 200


def _owned_chat(db: DBSession, chat_id: str, user_id: str) -> Chat:
    chat = db.get(Chat, chat_id)
    # 404 rather than 403 for someone else's chat: whether a given id exists is
    # not information this endpoint should hand out.
    if chat is None or chat.user_id != user_id:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That chat no longer exists.")
    return chat


def _new_chat(
    db: DBSession,
    user: User,
    *,
    folder_id: str = "",
    model_id: str = "",
    temporary: bool = False,
    kind: str = KIND_CHAT,
    ssh_profile_id: str = "",
    project_dir: str = "",
    agent_mode: str = "",
    reasoning_effort: str = "",
) -> Chat:
    """Create a chat row, resolving which model it should use.

    An agent chat's connection is settled here and never again. That is the
    lock: the harness, the tools offered and the approval loop all differ, so a
    conversation whose earlier turns ran somewhere else is not one conversation.

    The mode is *not* part of that lock and is accepted here so it can be chosen
    before the first word. Without it, reaching Plan mode meant starting a chat
    in Manual, sending something to make the chat exist, and only then being
    offered the control -- by which point the model had already answered under
    the wrong rules. The reasoning effort is accepted for the same reason, and
    wins over the model's default: an explicit choice beats an inherited one.
    """
    chosen = None
    if model_id:
        match = next(
            (m for m in chat_service.available_models(db, user) if m.model_id == model_id), None
        )
        if match is not None:
            chosen = (match.model_id, match.connection_id)
    if chosen is None:
        chosen = chat_service.default_model(db, user)

    # `Model.params_json` has said "default sampling params applied to new chats
    # using this model" since it was added and has been applied nowhere. It is
    # empty on every existing row, so honouring it now changes nothing until an
    # administrator sets something -- and it is what makes a per-model default
    # reasoning effort possible without a second column meaning the same thing.
    defaults: dict = {}
    if chosen is not None:
        model = db.scalar(
            select(Model).where(
                Model.model_id == chosen[0], Model.connection_id == chosen[1]
            )
        )
        if model is not None:
            defaults = dict(model.params_json or {})

    profile = _agent_target(db, user, kind, ssh_profile_id)
    chat = Chat(
        user_id=user.id,
        folder_id=folder_id or None,
        model_id=chosen[0] if chosen else "",
        connection_id=chosen[1] if chosen else None,
        temporary=temporary,
        kind=KIND_AGENT if profile is not None else KIND_CHAT,
        ssh_profile_id=profile.id if profile is not None else None,
        project_dir=(project_dir.strip() or profile.default_dir) if profile is not None else "",
        params_json=defaults,
    )
    # Ignored rather than refused when it is not a mode, matching how every
    # other bad value here collapses: somebody who mistypes should get a chat
    # under the safest rules, not an error page holding their message hostage.
    # Left alone entirely on a plain chat, where it means nothing.
    if profile is not None and agent_mode.strip() in agent_policy.MODES:
        chat.agent_mode = agent_mode.strip()
    # After the model's defaults, so choosing one on the new-chat screen wins
    # over the administrator's.
    #
    # `"off"` is a sentinel, and it has to be: `reasoning_effort` arrives as
    # `Form("")`, so an absent field and an empty one are indistinguishable --
    # the FastAPI trap this codebase has already been bitten by once. With
    # `value=""` on the off option, the reader would pick "off", the value would
    # fall out of EFFORTS, the model's default seeded above would stay, and they
    # would silently get "high". The picker shows what will be sent, so the two
    # have to agree.
    wanted_effort = reasoning_effort.strip().lower()
    if wanted_effort == "off":
        chat.params_json = {k: v for k, v in chat.params_json.items() if k != "reasoning_effort"}
    elif wanted_effort in chat_service.EFFORTS:
        chat.params_json = {**chat.params_json, "reasoning_effort": wanted_effort}
    db.add(chat)
    db.commit()
    return chat


@router.post("/start", dependencies=[Depends(require_permission("chat.create"))])
async def start_chat(
    db: Db,
    user: RequiredUser,
    content: str = Form(""),
    file_ids: list[str] = Form(default=[]),
    folder_id: str = Form(""),
    model_id: str = Form(""),
    temporary: bool = Form(False),
    kind: str = Form(KIND_CHAT),
    ssh_profile_id: str = Form(""),
    project_dir: str = Form(""),
    agent_mode: str = Form(""),
    reasoning_effort: str = Form(""),
) -> Response:
    """Create a chat from its first message.

    Chats are made here rather than by a "New chat" button so that an opened-
    and-abandoned chat never exists: the row appears only once there is
    something in it. The reply then streams the same way as any other, because
    /chat/{id} renders the unfinished assistant message with its sse-connect.
    """
    content = content.strip()
    if not content and not file_ids:
        return Response(status_code=status.HTTP_204_NO_CONTENT)

    chat = _new_chat(
        db,
        user,
        folder_id=folder_id,
        model_id=model_id,
        temporary=temporary,
        kind=kind,
        ssh_profile_id=ssh_profile_id,
        project_dir=project_dir,
        agent_mode=agent_mode,
        reasoning_effort=reasoning_effort,
    )

    user_message = chat_service.create_message(db, chat, ROLE_USER, content)
    if file_ids:
        files_service.claim(db, ids=file_ids, user_id=user.id, message_id=user_message.id)
    assistant = chat_service.create_message(
        db, chat, ROLE_ASSISTANT, "", complete_=False, model_id=chat.model_id
    )
    generation_service.ensure(chat.id, assistant.id)

    response = Response(status_code=status.HTTP_204_NO_CONTENT)
    response.headers["HX-Redirect"] = f"/chat/{chat.id}"
    return response


def _agent_target(db: DBSession, user: User, kind: str, profile_id: str):
    """The connection an agent chat is being pointed at, or None.

    Every "no" collapses to None and the chat is an ordinary one: not asked
    for, no permission, the feature off, or a profile that is not this person's.
    Refusing outright would be worse -- somebody whose permission was withdrawn
    between opening the composer and sending would lose the message.
    """
    from lembas.db.models import SshProfile
    from lembas.security import permissions

    if kind != KIND_AGENT or not profile_id:
        return None
    if not permissions.has(db, user, "tools.agent"):
        return None
    if not settings_store.agents(db).get("enabled"):
        return None

    profile = db.get(SshProfile, profile_id)
    # Ownership re-checked rather than trusted from the form: an id in a POST is
    # not an authorisation, and these are credentials to somebody's machine.
    if profile is None or profile.owner_id != user.id or not profile.enabled:
        return None
    return profile


# There is deliberately no route that creates an empty chat. Starting one is
# navigation to /chat (optionally ?model=...), and the row is written by
# /start when the first message is actually sent.


@router.get("/{chat_id}/inspect")
async def inspect_chat(request: Request, db: Db, user: RequiredUser, chat_id: str) -> Response:
    """What this chat would send upstream right now.

    Owner-checked *and* admin-checked, not admin alone. `permissions.resolve`
    giving an admin everything is about configuration, which they can grant
    themselves anyway; reading someone's conversation is a different act, which
    is why `sharing.visible_to` has no admin branch either. An inspector that
    could dump any user's transcript would be that branch under another name.

    Rebuilt, not recorded. Recording every request would store a copy of the
    whole conversation against every message, which grows quadratically with
    chat length -- and the thing an administrator actually wants to see is what
    the current configuration produces. The panel says so in as many words.
    """
    chat = _owned_chat(db, chat_id, user.id)
    if not user.is_admin:
        raise HTTPException(
            status.HTTP_403_FORBIDDEN, "The inspector is restricted to administrators."
        )

    offered = tools_service.enabled_tools(db, chat, user)
    payload = chat_service.build_request(db, chat, tools=offered, user=user)
    last = db.scalar(
        select(Message)
        .where(Message.chat_id == chat.id, Message.role == ROLE_ASSISTANT)
        .order_by(Message.created_at.desc())
    )

    messages = payload.get("messages") or []
    system = messages[0]["content"] if messages and messages[0].get("role") == "system" else ""

    return render(
        request,
        "chat/_inspector_body.html",
        {
            "chat": chat,
            "system": system,
            "request_json": _pretty(_redact(payload)),
            "row": last,
            "metrics": metrics_service.from_message(last.usage_json if last else None),
            "tool_names": [
                (t.get("function") or {}).get("name", "") for t in offered
            ],
            "model": chat_service.model_for(db, chat),
        },
    )


@router.get("/{chat_id}/usage")
async def chat_usage(request: Request, db: Db, user: RequiredUser, chat_id: str) -> Response:
    """What this conversation has cost, and how full the window is.

    Owner-checked and nothing else: it is your own chat's totals. Unlike the
    inspector next door there is no admin branch, because there is no reason
    for one -- the numbers describe a conversation, and reading somebody's
    conversation is exactly what `sharing` has no admin branch for either.

    Summed from what each reply recorded rather than recomputed: an endpoint
    that reported no usage contributed an estimate at the time, and re-deriving
    it now with a different estimator would make the totals move under a chat
    that had not changed.
    """
    chat = _owned_chat(db, chat_id, user.id)
    replies = list(
        db.scalars(
            select(Message)
            .where(Message.chat_id == chat.id, Message.role == ROLE_ASSISTANT)
            .order_by(Message.created_at)
        )
    )

    totals = {"prompt": 0, "completion": 0, "total": 0}
    estimated = False
    for reply in replies:
        usage = metrics_service.from_message(reply.usage_json)
        totals["prompt"] += usage.prompt_tokens
        totals["completion"] += usage.completion_tokens
        totals["total"] += usage.total_tokens
        estimated = estimated or usage.estimated

    last = replies[-1] if replies else None
    return render(
        request,
        "chat/_usage.html",
        {
            "chat": chat,
            "totals": totals,
            "estimated": estimated,
            "replies": len(replies),
            "metrics": metrics_service.from_message(last.usage_json if last else None),
            "model": chat_service.model_for(db, chat),
        },
    )


# Roughly what a downscaled phone photo comes to as base64. The exact figure
# does not matter; putting megabytes of it into the DOM does.
_REDACTED_URI = "data:…base64 image omitted…"
MAX_INSPECT_CHARS = 40_000


def _redact(payload: dict) -> dict:
    """Replace image data URIs before dumping.

    Nothing else is hidden -- fidelity is the whole point of the panel, and API
    keys never appear because `build_request` returns a body, not headers.
    """
    messages = []
    for message in payload.get("messages") or []:
        content = message.get("content")
        if isinstance(content, list):
            parts = []
            for part in content:
                if isinstance(part, dict) and part.get("type") == "image_url":
                    parts.append({"type": "image_url", "image_url": {"url": _REDACTED_URI}})
                else:
                    parts.append(part)
            message = {**message, "content": parts}
        messages.append(message)
    return {**payload, "messages": messages}


def _pretty(payload: dict) -> str:
    text = json.dumps(payload, indent=2, ensure_ascii=False, default=str)
    if len(text) > MAX_INSPECT_CHARS:
        return text[:MAX_INSPECT_CHARS] + "\n… truncated"
    return text


@router.post("/{chat_id}/compact")
async def compact_chat(request: Request, db: Db, user: RequiredUser, chat_id: str) -> Response:
    """Summarise the earlier turns and stop sending them.

    No permission of its own: compaction changes only what one chat sends
    upstream, and gating it would mean answering "why can this user not tidy
    their own conversation".
    """
    chat = _owned_chat(db, chat_id, user.id)

    unfinished = db.scalar(
        select(Message).where(Message.chat_id == chat.id, Message.complete.is_(False))
    )
    if unfinished is not None:
        # Summarising a transcript that is still being written races
        # build_request. Queuing it is a state machine nobody asked for.
        raise HTTPException(
            status.HTTP_409_CONFLICT, "Wait for the current reply to finish, then compact."
        )

    template = prompts_service.resolve(db, "task.compact")
    if not template.strip():
        raise HTTPException(
            status.HTTP_409_CONFLICT,
            "Compaction is turned off: its prompt is empty under Admin → Prompts.",
        )

    upto = compaction_service.last_complete(db, chat)
    if upto is None:
        raise HTTPException(
            status.HTTP_409_CONFLICT, "There is nothing here to summarise yet."
        )

    endpoint, model_id = chat_service.resolve_endpoint(db, chat)
    transcript = compaction_service.transcript(db, chat, upto=upto)
    previous = compaction_service.previous_summary_block(chat)

    summary = await chat_service.summarise_for_compaction(
        endpoint,
        model_id,
        transcript=transcript,
        previous_summary=previous,
        template=template,
    )
    if not summary:
        raise HTTPException(
            status.HTTP_409_CONFLICT, "The model returned no summary, so nothing changed."
        )

    compaction_service.apply(chat, summary=summary, upto=upto)
    db.commit()
    log.info("chat %s compacted through %s", chat.id, upto.id)

    return templates.TemplateResponse(
        request, "chat/_thread.html", {"request": request, **_thread_context(db, chat, user)}
    )


@router.post("/{chat_id}/index")
async def reindex_chat(db: Db, user: RequiredUser, chat_id: str) -> Response:
    """Walk the project directory again, now.

    The listing is cached for five minutes and only ever built when a reply
    starts, so a tree that has just changed under somebody's hands -- a checkout,
    a build, anything done in the terminal panel rather than through
    `file_write` -- stays wrong until the next reply after the TTL lapses. This
    is the "look again" that was missing.

    Read-only, and therefore outside `agent/policy.py` for the reason the
    directory browser is: it is LLeMbas acting on a person's instruction, not a
    model choosing to look, and a listing that asked permission would be
    useless. The gate is ownership of the connection, checked here rather than
    trusted from the chat.
    """
    from lembas.db.models import SshProfile
    from lembas.services.agent import index as index_service
    from lembas.services.agent import ssh as ssh_service

    chat = _owned_chat(db, chat_id, user.id)
    if chat.kind != KIND_AGENT or not chat.ssh_profile_id:
        raise HTTPException(status.HTTP_409_CONFLICT, "This chat has no project directory.")
    if not permissions.has(db, user, "tools.agent"):
        raise HTTPException(status.HTTP_403_FORBIDDEN, "You may not use agent connections.")

    profile = db.get(SshProfile, chat.ssh_profile_id)
    if profile is None or profile.owner_id != user.id or not profile.enabled:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That connection is not available.")
    if not profile.host_key:
        raise HTTPException(
            status.HTTP_409_CONFLICT,
            "This connection's host key has not been accepted yet.",
        )

    project_dir = chat.project_dir or profile.default_dir or ""
    try:
        found = await index_service.ensure(
            ssh_service.SshExecutor(ssh_service.spec_from(profile), project_dir),
            profile.id,
            project_dir,
            refresh=True,
        )
    except Exception as exc:  # noqa: BLE001 - surfaced to the reader, not swallowed
        log.warning("could not index %s for chat %s: %s", project_dir, chat.id, exc)
        raise HTTPException(
            status.HTTP_502_BAD_GATEWAY, "Could not read the project directory."
        ) from exc

    listed = len(found.paths)
    return JSONResponse(
        {
            "ok": True,
            "files": found.total,
            "listed": listed,
            "truncated": found.truncated,
            "message": (
                f"{found.total} files under {project_dir or '~'}"
                + (f", {listed} listed." if listed != found.total else ".")
            ),
        }
    )


@router.post("/{chat_id}/bases")
async def attach_base(
    request: Request, db: Db, user: RequiredUser, chat_id: str, base_id: str = Form("")
) -> Response:
    """Scope this chat to a knowledge base, from the `@` menu.

    A base is a *reference*, not an attachment: `Chat.knowledge_bases` already
    narrows `knowledge_search`, and the harness already names the attached bases
    so the model can tell "there is nothing about this" from "I can only see
    this folder". Copying a folder of documents into the window instead would
    cost the context on every request forever to answer one question.

    Additive, and idempotent -- choosing the same base twice is not an error.
    Removing one is a checkbox in the chat's settings, where the whole set is
    visible at once.
    """
    from lembas.db.models import KnowledgeBase
    from lembas.services.library import documents as documents_service

    chat = _owned_chat(db, chat_id, user.id)
    if not permissions.has(db, user, "library.use"):
        raise HTTPException(status.HTTP_403_FORBIDDEN, "You may not use the library.")

    base = db.scalar(
        documents_service.visible_bases(db, user).where(KnowledgeBase.id == base_id)
    )
    if base is None:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That knowledge base is not available.")

    if base.id not in {b.id for b in chat.knowledge_bases}:
        chat.knowledge_bases = [*chat.knowledge_bases, base]
        db.commit()

    return templates.TemplateResponse(
        request, "chat/_base_chip.html", {"request": request, "base": base}
    )


@router.post("/{chat_id}/scope")
async def set_scope(
    db: Db,
    user: RequiredUser,
    chat_id: str,
    kind: str = Form(""),
    name: str = Form(""),
    on: bool = Form(False),
) -> Response:
    """Turn one thing this chat may use on or off. **Narrowing only.**

    Nothing here widens anything. `resolve_tools` applies this *after* the
    model's capabilities, the reader's permissions and the instance
    configuration, so a crafted POST turning something on reaches a tool those
    gates have already removed -- there is a test for exactly that.

    On is stored by **removing** the key rather than by writing True, so absent
    stays the single representation of "on" and the column cannot grow a row per
    family per chat. Bounded, so a crafted request cannot grow it either.

    JSON reassignment rather than mutation: a plain dict assignment into a JSON
    column is not detected.
    """
    chat = _owned_chat(db, chat_id, user.id)
    bucket = {"family": "families", "skill": "skills"}.get(kind.strip())
    wanted = name.strip()[:64]
    if bucket is None or not wanted:
        raise HTTPException(status.HTTP_400_BAD_REQUEST, "Say what to turn on or off.")

    scope = dict(chat.scope_json or {})
    entries = dict(scope.get(bucket) or {})
    if on:
        entries.pop(wanted, None)
    else:
        if len(entries) >= MAX_SCOPE_KEYS:
            raise HTTPException(status.HTTP_409_CONFLICT, "Too many things switched off.")
        entries[wanted] = False

    if entries:
        scope[bucket] = entries
    else:
        scope.pop(bucket, None)
    chat.scope_json = scope
    db.commit()

    return Response(status_code=status.HTTP_204_NO_CONTENT)


@router.post("/{chat_id}/keep")
async def keep_chat(db: Db, user: RequiredUser, chat_id: str) -> Response:
    """Stop a temporary chat being temporary.

    A conversation that turns out to matter has to have a way out; without one,
    the sweep destroys it a day later with no recourse, and people discover that
    exactly once.
    """
    chat = _owned_chat(db, chat_id, user.id)
    chat.temporary = False
    db.commit()

    response = Response(status_code=status.HTTP_204_NO_CONTENT)
    # The sidebar has to gain a row and the topbar has to lose a badge; a full
    # refresh is one line against a handful of out-of-band fragments.
    response.headers["HX-Refresh"] = "true"
    return response


@router.get("/unread")
async def unread_poll(db: Db, user: RequiredUser) -> Response:
    """Dots for the sidebar, and a toast for anything newly arrived.

    Polled rather than pushed: a browser sitting on a different chat has no
    open connection to the one that finished, and a second always-on channel
    per tab is a lot of machinery for a green dot.

    Returns out-of-band spans so only the dots change -- re-rendering the whole
    sidebar would reset the folder open/closed state on every tick.
    """
    chats = list(
        db.scalars(
            select(Chat).where(
                Chat.user_id == user.id,
                Chat.archived.is_(False),
                # A temporary chat has no sidebar row, so a dot has nowhere to
                # land and the toast would name a chat nobody can navigate to.
                Chat.temporary.is_(False),
            )
        )
    )

    fresh = [c for c in chats if c.unread and not c.unread_notified]
    for chat in fresh:
        chat.unread_notified = True
    if fresh:
        db.commit()

    markup = "".join(
        f'<span id="unread-{c.id}" class="unread-dot" hx-swap-oob="true"'
        f'{"" if c.unread else " hidden"} title="New reply"></span>'
        for c in chats
    )

    response = HTMLResponse(markup)
    if fresh:
        # HX-Trigger carries the toast; ui.js listens for it.
        response.headers["HX-Trigger"] = json.dumps(
            {"lembas:unread": {"titles": [c.title for c in fresh]}}
        )
    return response


@router.post("/{chat_id}/messages")
async def post_message(
    request: Request,
    db: Db,
    user: RequiredUser,
    chat_id: str,
    content: str = Form(""),
    file_ids: list[str] = Form(default=[]),
) -> Response:
    """Persist the user's turn and hand back the pair of bubbles.

    The assistant bubble comes back empty, carrying the sse-connect attribute
    that opens the stream below. Splitting it this way means the POST returns
    immediately and the slow part is a separate, resumable connection.
    """
    chat = _owned_chat(db, chat_id, user.id)

    content = content.strip()
    # "Here, look at this" with no words is a legitimate turn, so an empty
    # message is only empty when it carries nothing at all.
    if not content and not file_ids:
        return Response(status_code=status.HTTP_204_NO_CONTENT)

    return _send(request, db, chat, user, content, file_ids=file_ids)


def _reply_in_flight(db: DBSession, chat: Chat) -> bool:
    """Whether this chat already has a reply being written.

    The row is the authority, not the registry: a restart leaves an incomplete
    assistant message behind with no `Generation` anywhere, and that row is what
    starts the reply again on the next page load. The registry is consulted too,
    for the sliver in which a generation is still running and its row has
    already been written -- `_persist` sets `complete` before `_run` sets
    `done`.
    """
    unfinished = db.scalar(
        select(Message).where(Message.chat_id == chat.id, Message.complete.is_(False))
    )
    return unfinished is not None or generation_service.running_for(chat.id) is not None


def _note_rewind(chat: Chat) -> None:
    """Record that an agent chat's transcript went back and the machine did not.

    Deliberately no attempt to undo anything out there. The project directory is
    somebody's real working tree, and deleting their work to match a rewound
    transcript would be far worse than the inconsistency. So the model is told
    instead -- see the `tool.agent_rewound` fragment -- and can look rather than
    assume.
    """
    if chat.kind == KIND_AGENT:
        chat.rewound_at = datetime.now(UTC)


def _send(
    request: Request,
    db: Db,
    chat: Chat,
    user: User,
    content: str,
    *,
    file_ids: list[str] | None = None,
) -> Response:
    """Write a turn, start the reply, and hand back the pair of bubbles.

    Shared by the composer and by anything else that puts words into a
    conversation on somebody's behalf -- carrying out a plan, for one. One path
    rather than two, so a second way of sending cannot drift from the first.

    If a reply is already being written, the turn is *queued* instead: written,
    shown, and not sent. Starting a second reply here is what used to happen,
    and it produced two generations answering the same chat from two different
    prefixes of it, with Stop pointing at whichever bubble came first in the
    document.
    """
    if queued := _reply_in_flight(db, chat):
        waiting = db.scalar(
            select(func.count())
            .select_from(Message)
            .where(Message.chat_id == chat.id, Message.queued.is_(True))
        )
        if waiting >= MAX_QUEUED:
            raise HTTPException(
                status.HTTP_409_CONFLICT,
                f"There are already {MAX_QUEUED} messages waiting to be sent.",
            )

    user_message = chat_service.create_message(db, chat, ROLE_USER, content, queued=queued)
    if file_ids:
        files_service.claim(db, ids=file_ids, user_id=user.id, message_id=user_message.id)
        db.refresh(user_message)

    if queued:
        # One bubble and no assistant placeholder. The streaming shell is the
        # only thing that starts a generation, so a placeholder here would be a
        # second concurrent reply -- exactly what the queue exists to prevent.
        if (live := generation_service.running_for(chat.id)) is not None:
            # So a reply between two rounds of tool calls notices it, and so
            # anybody following sees the status change.
            live.touch()
        return templates.TemplateResponse(
            request,
            "chat/_message.html",
            {
                "request": request,
                "message": user_message,
                "chat": chat,
                "user": user,
                "models_by_id": {
                    m.model_id: m for m in chat_service.available_models(db, user)
                },
                **audio_service.template_flags(db, user),
            },
        )

    assistant_message = chat_service.create_message(
        db, chat, ROLE_ASSISTANT, "", complete_=False, model_id=chat.model_id
    )
    generation_service.ensure(chat.id, assistant_message.id)

    # `user` is required by the shared message template, which renders both
    # roles; without it the user bubble's initial blows up.
    return templates.TemplateResponse(
        request,
        "chat/_turn.html",
        {
            "request": request,
            "user_message": user_message,
            "assistant_message": assistant_message,
            "chat": chat,
            "user": user,
            "models_by_id": {
                m.model_id: m for m in chat_service.available_models(db, user)
            },
            **audio_service.template_flags(db, user),
        },
    )


@router.get("/{chat_id}/messages/{message_id}/stream")
async def stream_message(
    db: Db,
    user: RequiredUser,
    chat_id: str,
    message_id: str,
) -> Response:
    """Stream the assistant's reply as server-sent events.

    Emits `token` events carrying escaped text, then a single `done` event
    carrying the finished bubble rendered from Markdown, then `close`.
    """
    chat = _owned_chat(db, chat_id, user.id)
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat.id:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That message no longer exists.")

    return StreamingResponse(
        _follow(chat.id, message.id),
        media_type="text/event-stream",
        headers={
            "Cache-Control": "no-cache, no-transform",
            "Connection": "keep-alive",
            # nginx buffers proxied responses by default, which turns a stream
            # into one delivery at the end. This is the documented opt-out.
            "X-Accel-Buffering": "no",
        },
    )


def _tool_activity(events: list[dict], *, live: bool = True) -> str:
    """Render the tool block. Whole, never a delta, like every other frame."""
    return templates.get_template("chat/_tool_activity.html").render(
        {"tool_events": events, "live": live}
    )


def _ask_html(chat_id: str, pending) -> str:
    """The card asking the reader something, or nothing at all.

    Returns "" when there is nothing pending, and the frame is sent
    unconditionally, because this is one of the few blocks that has to be able
    to *clear* itself: the card must vanish the moment it is answered.
    `reasoning`, `tools` and `render` are the opposite -- guarded by truthiness
    so a frame can never blank them.
    """
    if pending is None:
        return ""
    return templates.get_template("chat/_interaction.html").render(
        {"ask": pending, "chat_id": chat_id}
    )


async def _follow(chat_id: str, message_id: str) -> AsyncIterator[str]:
    """Stream a generation that is running independently of this request.

    This connection only *watches*. Closing it -- navigating away, opening
    another chat -- leaves the reply being written, and reconnecting replays
    the whole state immediately rather than starting over.

    Both `render` and `reasoning` carry the complete block each time rather
    than a delta, which is what makes reattaching mid-reply work at all: a
    follower arriving late has no earlier fragments to append to.
    """
    generation = generation_service.ensure(chat_id, message_id)
    generation.followers += 1
    seen = -1
    last_frame = time.monotonic()

    try:
        while True:
            if generation.version != seen:
                seen = generation.version
                if generation.thinking:
                    yield sse.event("reasoning", escape_text(generation.thinking))
                if generation.tool_events:
                    yield sse.event("tools", _tool_activity(generation.tool_events))
                if generation.content:
                    yield sse.event("render", render_markdown(generation.text))
                yield sse.event("metrics", _metrics_html(generation))
                yield sse.event("status", escape_text(generation.status))
                yield sse.event("ask", _ask_html(chat_id, generation.pending))
                last_frame = time.monotonic()

            if generation.done:
                break

            # A reasoning model can think for a minute or more without emitting
            # anything, and an idle connection is what a proxy closes. The
            # comment frame keeps it open and is ignored by the browser.
            if time.monotonic() - last_frame > KEEPALIVE_AFTER:
                yield sse.KEEPALIVE
                last_frame = time.monotonic()

            # Polling rather than per-follower wakeups: the producer already
            # works in RENDER_INTERVAL steps, so a short sleep is simpler and
            # cannot drop a notification.
            await asyncio.sleep(generation_service.RENDER_INTERVAL * 0.8)
    finally:
        generation.followers = max(0, generation.followers - 1)

    # The producer commits the message before marking itself done, so by here
    # the row is authoritative and the final bubble can be rendered from it.
    with session_scope() as db:
        message = db.get(Message, message_id)
        chat = db.get(Chat, chat_id)
        if message is None or chat is None:
            yield sse.event("close", "")
            return

        owner = db.get(User, chat.user_id)
        final_html = templates.get_template("chat/_message.html").render(
            {
                "message": message,
                "body_html": render_markdown(message.content),
                "chat": chat,
                # Passed even though an assistant bubble never reads it: the
                # template shares both roles, and a missing `user` would only
                # blow up on whichever branch is not being exercised here.
                "user": owner,
                "models_by_id": {
                    m.model_id: m for m in chat_service.available_models(db, None)
                },
                # This frame replaces the whole bubble, so it has to carry the
                # speaker button's conditions too -- and the owner's, not the
                # follower's: there is no request here to ask who is watching.
                **audio_service.template_flags(db, owner),
                # The one render that means "this reply just landed", which is
                # what read-aloud-automatically keys off. A page load must not
                # set it or reopening a chat would start talking.
                "just_finished": True,
            }
        )
        title_html = templates.get_template("chat/_title_oob.html").render({"chat": chat})

        # What the queue did while this reply was running. There is no push
        # channel that outlives one message's stream, and this is the last frame
        # that reaches the browser -- so it carries the rest out of band, the
        # way the chat title already does.
        moved_html, queue_html = _queue_frames(db, chat, owner, generation)

    yield sse.event("done", moved_html + final_html + queue_html + title_html)
    yield sse.event("close", "")


def _render_bubble(db: DBSession, chat: Chat, owner: User | None, message: Message) -> str:
    """One finished bubble, rendered the way the `done` frame renders its own."""
    return templates.get_template("chat/_message.html").render(
        {
            "message": message,
            "body_html": (
                render_markdown(message.content) if message.role == ROLE_ASSISTANT else ""
            ),
            "chat": chat,
            "user": owner,
            "models_by_id": {m.model_id: m for m in chat_service.available_models(db, None)},
            **audio_service.template_flags(db, owner),
        }
    )


def _queue_frames(
    db: DBSession, chat: Chat, owner: User | None, generation
) -> tuple[str, str]:
    """The bubbles the queue produced during this reply, as out-of-band HTML.

    Two pieces, because they swap differently. Anything taken *into* this reply
    mid-round now sorts before it, so it is rendered ahead of the finished
    bubble in the same `outerHTML` swap and its stale node is deleted out of
    band -- one frame, and the DOM ends up in the order the database is in.
    Anything drained *after* the reply is a new pair appended to the thread.
    """
    moved: list[str] = []
    out_of_band: list[str] = []

    for injected_id in generation.injected_ids:
        row = db.get(Message, injected_id)
        if row is None:
            continue
        moved.append(_render_bubble(db, chat, owner, row))
        # Removed where it was; it is about to reappear above the reply.
        out_of_band.append(f'<article id="msg-{row.id}" hx-swap-oob="delete"></article>')

    if generation.drained:
        fresh = list(
            db.scalars(
                select(Message)
                .where(Message.chat_id == chat.id, Message.complete.is_(False))
                .order_by(Message.created_at)
            )
        )
        for assistant in fresh:
            # The user turn that was waiting has just lost its Send now and
            # Discard, so it is re-rendered in place.
            delivered = db.scalars(
                select(Message)
                .where(
                    Message.chat_id == chat.id,
                    Message.role == ROLE_USER,
                    Message.created_at <= assistant.created_at,
                )
                .order_by(Message.created_at.desc())
                .limit(1)
            ).first()
            if delivered is not None:
                out_of_band.append(
                    f'<div hx-swap-oob="outerHTML:#msg-{delivered.id}">'
                    + _render_bubble(db, chat, owner, delivered)
                    + "</div>"
                )
            out_of_band.append(
                '<div hx-swap-oob="beforeend:#thread">'
                + _render_bubble(db, chat, owner, assistant)
                + "</div>"
            )

    return "".join(moved), "".join(out_of_band)


def _metrics_html(generation) -> str:
    """The metric chips for a reply still being written.

    Built from the same Metrics object the finished bubble uses, so the numbers
    do not jump when the stream ends -- the only thing that changes is that an
    estimate may have become exact.
    """
    return templates.get_template("chat/_metrics.html").render(
        {"metrics": metrics_service.from_generation(generation)}
    )


def _thread_context(db: DBSession, chat: Chat, user: User) -> dict:
    """Everything chat/_thread.html needs to render the conversation."""
    everything = list(
        db.scalars(select(Message).where(Message.chat_id == chat.id).order_by(Message.created_at))
    )
    compacted, messages = compaction_service.split(db, chat, everything)
    return {
        "chat": chat,
        "user": user,
        "messages": messages,
        "compacted": compacted,
        "bodies": {
            m.id: render_markdown(m.content)
            for m in everything
            if m.role == ROLE_ASSISTANT and m.content
        },
        "models_by_id": {m.model_id: m for m in chat_service.available_models(db, user)},
        **audio_service.template_flags(db, user),
    }


def _messages_after(db: DBSession, message: Message) -> list[Message]:
    return list(
        db.scalars(
            select(Message)
            .where(Message.chat_id == message.chat_id, Message.created_at > message.created_at)
            .order_by(Message.created_at)
        )
    )


@router.get("/{chat_id}/messages/{message_id}/edit")
async def edit_form(
    request: Request, db: Db, user: RequiredUser, chat_id: str, message_id: str
) -> Response:
    """Swap one of the reader's own turns into an editable form."""
    chat = _owned_chat(db, chat_id, user.id)
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat.id or message.role != ROLE_USER:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That message no longer exists.")

    return templates.TemplateResponse(
        request,
        "chat/_edit_form.html",
        {
            "request": request,
            "chat": chat,
            "user": user,
            "message": message,
            "following": len(_messages_after(db, message)),
        },
    )


@router.get("/{chat_id}/messages/{message_id}/cancel-edit")
async def cancel_edit(
    request: Request, db: Db, user: RequiredUser, chat_id: str, message_id: str
) -> Response:
    """Put the bubble back, unchanged."""
    chat = _owned_chat(db, chat_id, user.id)
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat.id:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That message no longer exists.")

    return templates.TemplateResponse(
        request,
        "chat/_message.html",
        {
            "request": request,
            "chat": chat,
            "user": user,
            "message": message,
            "body_html": "",
            "models_by_id": {},
        },
    )


@router.post("/{chat_id}/messages/{message_id}/edit")
async def edit_message(
    request: Request,
    db: Db,
    user: RequiredUser,
    chat_id: str,
    message_id: str,
    content: str = Form(...),
) -> Response:
    """Rewrite one of the reader's turns and run the conversation on from there.

    Everything after the edited message is deleted rather than branched. A
    branch would need a UI for choosing between versions, and "go back and try
    again from here" is what was actually asked for -- the simpler behaviour is
    also the one people expect from every other chat client.
    """
    chat = _owned_chat(db, chat_id, user.id)
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat.id or message.role != ROLE_USER:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That message no longer exists.")

    content = content.strip()
    if not content and not message.attachments:
        raise HTTPException(status.HTTP_400_BAD_REQUEST, "A message cannot be empty.")

    # Editing rewinds and then starts a reply, unconditionally. Doing that while
    # one is already being written is a second concurrent generation -- the
    # thing the queue exists to prevent -- reachable here by a button that is on
    # screen throughout. It was reachable before the queue too; nothing made it
    # obvious.
    if _reply_in_flight(db, chat):
        raise HTTPException(
            status.HTTP_409_CONFLICT, "Wait for the current reply to finish, or stop it."
        )

    message.content = content

    # Attachments cascade with their message, so the files go too.
    discarded = _messages_after(db, message)
    for later in discarded:
        db.delete(later)

    # A rewind to at or before the compaction boundary leaves that boundary
    # describing turns that no longer exist. There is no foreign key to null it
    # out on an upgraded database, so it is cleared here.
    cutoff = compaction_service.cutoff_message(db, chat)
    if cutoff is None or compaction_service.moment(message) <= compaction_service.moment(cutoff):
        compaction_service.reset(chat)

    _note_rewind(chat)
    db.commit()

    assistant = chat_service.create_message(
        db, chat, ROLE_ASSISTANT, "", complete_=False, model_id=chat.model_id
    )
    generation_service.ensure(chat.id, assistant.id)
    log.info("chat %s rewound to a message, %d discarded", chat.id, len(discarded))

    return templates.TemplateResponse(
        request, "chat/_thread.html", {"request": request, **_thread_context(db, chat, user)}
    )


@router.post("/{chat_id}/messages/{message_id}/execute-plan")
async def execute_plan(
    request: Request, db: Db, user: RequiredUser, chat_id: str, message_id: str
) -> Response:
    """Carry out a plan the model proposed.

    Switches to **Edit**, never Auto. The plan was written under a mode where
    every command stopped for approval, and a button that also removed the
    asking is not the button anybody pressed.

    The plan is sent back **marked as a quotation of the model's own words**
    rather than as a bare instruction. A plan whose text came out of a file the
    model read would otherwise arrive in the most trusted role in the
    transcript, wearing the reader's authority -- which is precisely how an
    injected instruction would like to arrive.
    """
    chat = _owned_chat(db, chat_id, user.id)
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat.id or not message.plan_json:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "There is no plan on that message.")
    if chat.kind != KIND_AGENT:
        raise HTTPException(status.HTTP_409_CONFLICT, "This chat cannot act on anything.")

    # `message.plan`, the property, so a row written before version 2 comes
    # through as one phase. `steps` is flattened from every phase in order and
    # is always written, which is why this line needed no change when the shape
    # grew findings, objectives and phases.
    plan = message.plan
    steps = [str(s) for s in (plan.get("steps") or [])]
    body = "\n".join(f"{n}. {step}" for n, step in enumerate(steps, start=1))

    chat.agent_mode = agent_policy.MODE_EDIT
    # The chat is now working to this plan, so the harness puts it in front of
    # the model each turn and `plan_update` is offered. Without this the model
    # carrying it out cannot see the plan it is carrying out, and could not tick
    # anything off if it wanted to.
    chat.plan_message_id = message.id
    db.commit()

    content = (
        "Carry out the plan you proposed above:\n\n"
        f"> **{plan.get('title') or 'The plan'}**\n"
        + "\n".join(f"> {line}" for line in body.splitlines())
        + "\n\nWork through it in order. If a step turns out to be wrong, stop "
        "and say so rather than improvising around it."
    )
    log.info("%s executing a plan in chat %s", user.email, chat.id)
    return _send(request, db, chat, user, content)


@router.post("/{chat_id}/messages/{message_id}/stop")
async def stop_message(db: Db, user: RequiredUser, chat_id: str, message_id: str) -> Response:
    """Ask a running generation to stop.

    Whatever has arrived is kept: a half-written answer the reader chose to cut
    short is still worth having, and discarding it would be a surprise.
    """
    chat = _owned_chat(db, chat_id, user.id)
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat.id:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That message no longer exists.")

    generation_service.request_stop(message.id)
    return Response(status_code=status.HTTP_204_NO_CONTENT)


def _waiting_message(db: DBSession, chat: Chat, message_id: str) -> Message:
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat.id or not message.queued:
        raise HTTPException(
            status.HTTP_404_NOT_FOUND, "That message is not waiting to be sent."
        )
    return message


@router.post("/{chat_id}/messages/{message_id}/discard")
async def discard_queued(db: Db, user: RequiredUser, chat_id: str, message_id: str) -> Response:
    """Withdraw a prompt that has not been sent.

    Deleted outright rather than marked: it never reached a model, nothing in
    the transcript refers to it, and a conversation full of tombstones for
    things nobody said is worse than the row being gone. Attachments cascade.

    An empty body rather than a 204, because htmx does not swap on a 204 and the
    bubble has to disappear.
    """
    chat = _owned_chat(db, chat_id, user.id)
    message = _waiting_message(db, chat, message_id)

    db.delete(message)
    db.commit()
    return HTMLResponse("")


@router.post("/{chat_id}/messages/{message_id}/send-now")
async def send_queued_now(
    request: Request, db: Db, user: RequiredUser, chat_id: str, message_id: str
) -> Response:
    """Deliver a waiting prompt at once.

    Refused while a reply is being written rather than allowed to jump ahead of
    it: that is what the queue *is*, and starting a second generation here is
    the thing this whole mechanism exists to stop. Stop the reply first.

    The whole thread comes back, which is the rewind and compaction idiom, and
    is safe only because of the refusal above -- there is no live bubble to
    destroy.
    """
    chat = _owned_chat(db, chat_id, user.id)
    message = _waiting_message(db, chat, message_id)
    if _reply_in_flight(db, chat):
        raise HTTPException(
            status.HTTP_409_CONFLICT, "Wait for the current reply to finish, or stop it."
        )

    message.queued = False
    db.commit()
    assistant = chat_service.create_message(
        db, chat, ROLE_ASSISTANT, "", complete_=False, model_id=chat.model_id
    )
    generation_service.ensure(chat.id, assistant.id)

    return templates.TemplateResponse(
        request, "chat/_thread.html", {"request": request, **_thread_context(db, chat, user)}
    )


@router.post("/{chat_id}/interaction/{interaction_id}")
async def answer_interaction(
    request: Request,
    db: Db,
    user: RequiredUser,
    chat_id: str,
    interaction_id: str,
) -> Response:
    """Answer the questions, or allow the action, a reply is waiting on.

    The whole card comes back at once, which is why the raw form is read rather
    than declared parameters: one `ask_user` call may have put four questions,
    and each carries a chosen option and a box to write something else. Per
    question, what was written wins over what was picked -- somebody who typed
    in the box after clicking an option meant the typing.

    `_owned_chat` is the authorisation and it is not decoration: without it any
    signed-in account that guessed an id would be answering -- and later,
    approving a command in -- somebody else's conversation.

    An id matching nothing (already answered, timed out, or the server was
    restarted) is a 204 with a toast rather than a 404. The card is gone either
    way, and an error page swapped into the middle of a chat is worse than
    being told plainly.
    """
    chat = _owned_chat(db, chat_id, user.id)
    form = await request.form()

    answers: dict[str, str] = {}
    for field, value in form.multi_items():
        kind, _, key = str(field).partition(".")
        if not key or kind not in ("choice", "text"):
            continue
        written = str(value).strip()
        if kind == "text" and written:
            answers[key] = written
        elif kind == "choice" and written:
            answers.setdefault(key, written)

    answered = generation_service.answer(
        chat.id,
        interaction_id,
        verdict=str(form.get("verdict") or "").strip(),
        answers=answers,
    )

    response = Response(status_code=status.HTTP_204_NO_CONTENT)
    if not answered:
        response.headers["HX-Trigger"] = json.dumps(
            {"lembas:notify": {"message": "That question is no longer waiting for an answer."}}
        )
    return response


@router.patch("/{chat_id}")
async def update_chat(request: Request, db: Db, user: RequiredUser, chat_id: str) -> Response:
    """Partially update a chat.

    The raw form is read rather than declaring Form() parameters because
    FastAPI substitutes the default for an empty form value, which makes
    "field absent" and "field submitted empty" indistinguishable. That
    difference is exactly what this endpoint needs: an empty system prompt or
    temperature means *clear it*, not *leave it alone*.
    """
    chat = _owned_chat(db, chat_id, user.id)
    allowed = permissions.resolve(db, user)
    form = await request.form()

    if "title" in form:
        cleaned = str(form["title"]).strip()[:300]
        if cleaned:
            chat.title = cleaned
            # An explicit rename must not be overwritten by auto-titling later.
            chat.title_generated = True

    if "folder_id" in form:
        chat.folder_id = str(form["folder_id"]) or None

    # The mode is the one agent field that changes mid-chat: it decides what
    # gets asked about, not what the conversation is.
    if "agent_mode" in form:
        wanted = str(form["agent_mode"]).strip()
        if wanted in agent_policy.MODES:
            chat.agent_mode = wanted

    # And these are the ones that never do. Refused rather than ignored: a form
    # that quietly did nothing would look like a bug from the outside, and
    # without the refusal a crafted POST would repoint a conversation at another
    # machine halfway through.
    for locked in ("kind", "ssh_profile_id", "project_dir"):
        if locked in form:
            raise HTTPException(
                status.HTTP_409_CONFLICT,
                "A chat's connection is fixed when it is created. Start a new "
                "chat to work somewhere else.",
            )

    model_id = str(form.get("model_id", "")).strip()

    if model_id:
        if not allowed.get("chat.model_select"):
            raise HTTPException(
                status.HTTP_403_FORBIDDEN, "You may not change the model for a chat."
            )
        # Checked against what this user can reach, not merely what exists --
        # otherwise the picker is advisory and a crafted request bypasses it.
        match = next(
            (m for m in chat_service.available_models(db, user) if m.model_id == model_id),
            None,
        )
        if match is None:
            raise HTTPException(status.HTTP_403_FORBIDDEN, "That model is not available to you.")
        chat.model_id = model_id
        chat.connection_id = match.connection_id

    if "system_prompt" in form:
        if not allowed.get("chat.system_prompt"):
            raise HTTPException(
                status.HTTP_403_FORBIDDEN, "You may not set a system prompt."
            )
        chat.system_prompt = str(form["system_prompt"]).strip()[:8000]

    if "knowledge_base_ids" in form:
        # Sent as a single field even when empty, so that clearing every box
        # actually clears the attachment -- absent checkboxes carry no signal of
        # their own, which is the same trap update_chat exists to avoid.
        from lembas.db.models import KnowledgeBase
        from lembas.services.library import documents as documents_service

        wanted = [value for value in form.getlist("knowledge_base_ids") if value]
        chat.knowledge_bases = (
            list(
                db.scalars(
                    documents_service.visible_bases(db, user).where(
                        KnowledgeBase.id.in_(wanted)
                    )
                )
            )
            if wanted
            else []
        )

    submitted_params = {name: form[name] for name in _PARAM_RANGES if name in form}
    if submitted_params:
        if not allowed.get("chat.params"):
            raise HTTPException(
                status.HTTP_403_FORBIDDEN, "You may not change sampling parameters."
            )
        chat.params_json = {
            **(chat.params_json or {}),
            **_clean_params(**{k: str(v) for k, v in submitted_params.items()}),
        }

    # Not a number, so it cannot go through _PARAM_RANGES. `"off"` and empty
    # both clear it -- the sentinel because that is what the picker sends now,
    # empty because anything still posting the old value must keep working.
    # Anything that is neither is ignored rather than refused, so a typo does
    # not cost a message.
    if "reasoning_effort" in form:
        if not allowed.get("chat.params"):
            raise HTTPException(
                status.HTTP_403_FORBIDDEN, "You may not change sampling parameters."
            )
        wanted = str(form["reasoning_effort"]).strip().lower()
        if not wanted or wanted == "off":
            chat.params_json = {**(chat.params_json or {}), "reasoning_effort": None}
        elif wanted in chat_service.EFFORTS:
            chat.params_json = {**(chat.params_json or {}), "reasoning_effort": wanted}

    # Switching model re-seeds an effort that was never chosen, so "what the
    # picker shows is what is sent" stays true afterwards. Only when the key is
    # ABSENT: `None` means somebody cleared it deliberately, and resurrecting
    # that would make "off" silently do nothing on the next model change.
    if model_id and "reasoning_effort" not in (chat.params_json or {}):
        seeded = ((match.params_json if match is not None else None) or {}).get(
            "reasoning_effort"
        )
        if seeded in chat_service.EFFORTS:
            chat.params_json = {**(chat.params_json or {}), "reasoning_effort": seeded}

    db.commit()
    return Response(status_code=status.HTTP_204_NO_CONTENT)


# Bounds are the ones every provider agrees on. Out-of-range values are
# dropped rather than clamped: silently changing what someone typed is worse
# than ignoring it, and the form shows what actually stuck on reload.
_PARAM_RANGES: dict[str, tuple[type, float, float]] = {
    "temperature": (float, 0.0, 2.0),
    "top_p": (float, 0.0, 1.0),
    "max_tokens": (int, 1, 1_000_000),
}


def _clean_params(**submitted: str | None) -> dict[str, float | int | None]:
    """Parse sampling parameters, dropping anything unusable.

    An empty string means "unset this and let the provider default apply", so
    it maps to None rather than being ignored.
    """
    cleaned: dict[str, float | int | None] = {}
    for name, raw in submitted.items():
        if raw is None:
            continue
        if not raw.strip():
            cleaned[name] = None
            continue
        caster, low, high = _PARAM_RANGES[name]
        try:
            value = caster(raw)
        except (TypeError, ValueError):
            continue
        if low <= value <= high:
            cleaned[name] = value
    return cleaned


@router.delete("/{chat_id}", dependencies=[Depends(require_permission("chat.delete"))])
async def delete_chat(db: Db, user: RequiredUser, chat_id: str) -> Response:
    chat = _owned_chat(db, chat_id, user.id)
    # Before the row goes: a terminal is keyed on the chat id, so afterwards
    # there would be nothing left to find it by and a shell would sit open on
    # somebody's machine until the idle timeout noticed.
    await terminal_service.close_chat(chat_id)
    db.delete(chat)
    db.commit()

    response = Response(status_code=status.HTTP_204_NO_CONTENT)
    response.headers["HX-Redirect"] = "/chat"
    return response


@router.get("/{chat_id}/messages/{message_id}/raw")
async def raw_message(db: Db, user: RequiredUser, chat_id: str, message_id: str) -> HTMLResponse:
    """The unrendered Markdown of a message, for the copy button."""
    _owned_chat(db, chat_id, user.id)
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat_id:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That message no longer exists.")
    return HTMLResponse(escape_text(message.content))


@router.post("/{chat_id}/messages/{message_id}/regenerate")
async def regenerate(
    request: Request,
    db: Db,
    user: RequiredUser,
    chat_id: str,
    message_id: str,
) -> Response:
    """Discard an assistant reply and produce a fresh one in its place."""
    chat = _owned_chat(db, chat_id, user.id)
    message = db.get(Message, message_id)
    if message is None or message.chat_id != chat.id or message.role != ROLE_ASSISTANT:
        raise HTTPException(status.HTTP_404_NOT_FOUND, "That reply no longer exists.")

    message.content = ""
    message.error = ""
    message.complete = False
    message.model_id = chat.model_id
    _note_rewind(chat)
    db.commit()
    # restart, not ensure: this is the one caller that reuses a Message row, and
    # the finished generation for it is still registered.
    generation_service.restart(chat.id, message.id)

    return templates.TemplateResponse(
        request,
        "chat/_message.html",
        {
            "request": request,
            "message": message,
            "chat": chat,
            "body_html": "",
            "user": user,
            "models_by_id": {
                m.model_id: m for m in chat_service.available_models(db, user)
            },
            **audio_service.template_flags(db, user),
        },
    )


__all__ = ["render", "router"]