Compaction: a button, and automatically when the window fills

A long conversation eventually just stops working. Compaction summarises the
earlier turns and sends the summary in their place.

The messages are kept. They stay in the transcript behind a collapsed
divider and simply stop being part of the request, which is what makes the
button safe to press and automatic compaction safe to have at all: a summary
that came out badly is a bad turn, not a lost conversation.

Stored on the Chat, not as a synthetic Message. A synthetic row needs a
role -- `system` breaks the one-system-message rule the moment build_messages
emits it beside the harness, and user/assistant makes it a turn people can
edit, regenerate from and copy, indistinguishable from a real one in all
four places a bubble is rendered. Worse, "editing rewinds, it does not
branch" would silently delete it and leave no marker that compaction had
happened at all.

The summary goes out as a user turn and an assistant turn, not one. A
leading assistant breaks templates requiring the first non-system message to
be user; a lone leading user produces user, user whenever the kept history
starts on a user turn -- which it always does, because the cutoff lands on a
finished reply.

compacted_through_id is a plain id rather than a foreign key: migrations.py
compiles only the column type, so a REFERENCES clause would exist on a fresh
database and not on an upgraded one, and a constraint half the fleet has is
worse than none. cutoff_message validates it on every read instead, and a
rewind past the boundary clears it.

Compacting again summarises only the delta, with the previous summary
supplied to be subsumed. Re-summarising the whole chat each time grows
quadratically and eventually exceeds the window it is protecting.

Automatically at the top of _run, not in post_message: that route's contract
is to return immediately and leave the slow part to a resumable connection,
and it also means build_request is called once, after compaction, with no
second assembly path. The trigger is the last reply's recorded usage plus an
estimate of the new turn -- retrospective because true prompt_tokens are only
knowable after a response, plus the delta because otherwise fifty thousand
characters pasted into the composer overflow a window that read 90% last
turn. It never fires when the context length is unknown. It does fire on
estimated counts, which is safe here precisely because nothing is lost.

_maybe_compact never raises: a failure logs and sends the uncompacted
request. A `status` event says "Summarising earlier messages…" in the
meantime, because a silent multi-second pause before the first token is what
a hang looks like.

The wording is three fragments under Admin - Prompts. Clearing task.compact
turns compaction off entirely.

Also adds compaction.moment(): SQLite does not store the offset, so a row
loaded from disk is naive while one in the session's identity map keeps its
tzinfo, and comparing the two raises. Every comparison here is between
exactly those.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

src/lembas/services/chat.py

@@ -174,11 +174,31 @@ def build_messages(
     resolved, which is how the harness gets in front of the authored prompt
     without this function knowing anything about tools.
     """
+    from lembas.services import compaction as compaction_service
+    from lembas.services import prompts as prompts_service
+
     payload: list[dict[str, Any]] = []
     system = effective_system_prompt(db, chat) if system_prompt is None else system_prompt
     if system:
         payload.append({"role": ROLE_SYSTEM, "content": system})
 
+    # Compacted turns are replaced by a summary carried in two turns rather than
+    # one. A leading `assistant` breaks templates that require the first
+    # non-system message to be `user`; a lone leading `user` produces user, user
+    # whenever the kept history starts on a user turn -- which it always does,
+    # because the cutoff lands on a finished reply. The pair alternates
+    # correctly in both directions and keeps exactly one system message.
+    cutoff = compaction_service.cutoff_message(db, chat)
+    if cutoff is not None:
+        lead = prompts_service.resolve(db, "task.compact_lead").strip()
+        ack = prompts_service.resolve(db, "task.compact_ack").strip()
+        summary = chat.compact_summary.strip()
+        payload.append(
+            {"role": ROLE_USER, "content": f"{lead}\n\n{summary}" if lead else summary}
+        )
+        if ack:
+            payload.append({"role": ROLE_ASSISTANT, "content": ack})
+
     history = db.scalars(
         select(Message).where(Message.chat_id == chat.id).order_by(Message.created_at)
     ).all()
@@ -186,6 +206,10 @@ def build_messages(
     for message in history:
         if upto is not None and message.id == upto.id:
             break
+        if cutoff is not None and compaction_service.moment(
+            message
+        ) <= compaction_service.moment(cutoff):
+            continue
         # Skip turns that failed or produced nothing -- but a message carrying
         # only an attachment has no text and must still be sent.
         if message.error:
@@ -391,6 +415,42 @@ def create_message(
     return message
 
 
+async def summarise_for_compaction(
+    endpoint: Endpoint,
+    model_id: str,
+    *,
+    transcript: str,
+    previous_summary: str,
+    template: str,
+) -> str:
+    """Ask the model to summarise the earlier turns.
+
+    `template` is passed in for the same reason `generate_title`'s is: this runs
+    after the generation's session has closed, and opening another one there is
+    how you get a session that outlives its scope. An empty template means an
+    administrator cleared the fragment, and nothing is asked of anyone.
+    """
+    from lembas.services import prompts as prompts_service
+
+    if not template.strip() or not transcript.strip():
+        return ""
+
+    prompt = prompts_service.substitute(
+        template, {"transcript": transcript, "previous_summary": previous_summary}
+    )
+    raw = await complete(
+        endpoint,
+        {
+            "model": model_id,
+            "messages": [{"role": ROLE_USER, "content": prompt}],
+            "max_tokens": 1200,
+            # Low, but not zero: this is recall, not invention.
+            "temperature": 0.3,
+        },
+    )
+    return raw.strip()
+
+
 def sweep_temporary(db: DBSession, older_than: timedelta = TEMPORARY_LIFETIME) -> int:
     """Delete temporary chats nobody has touched for a day.
 

src/lembas/services/compaction.py

@@ -0,0 +1,192 @@
+"""Carrying a long conversation forward without carrying all of it.
+
+Past a certain length every chat stops working: the window fills, and the only
+options are to lose the beginning or to start again. Compaction summarises the
+earlier turns and sends the summary in their place.
+
+**The messages are kept.** They stay in the transcript, collapsed behind a
+divider, and simply stop being part of the request. A summary that turned out
+badly is then a bad turn rather than a lost conversation, which is what makes
+the button safe to press and automatic compaction safe to have at all.
+
+**Stored on the Chat, not as a synthetic Message.** A synthetic row would need a
+role: `system` breaks the one-system-message rule the moment `build_messages`
+emits it beside the harness, and `user`/`assistant` makes it a turn people can
+edit, regenerate from and copy, indistinguishable from a real one in all four
+places a bubble is rendered. Worse, "editing rewinds, it does not branch" would
+silently delete it and leave no marker that compaction had ever happened.
+"""
+
+from __future__ import annotations
+
+import logging
+from datetime import UTC, datetime
+
+from sqlalchemy import select
+from sqlalchemy.orm import Session as DBSession
+
+from lembas.db.models import ROLE_ASSISTANT, Chat, Message
+from lembas.services import metrics as metrics_service
+from lembas.services import settings_store, tokens
+
+log = logging.getLogger(__name__)
+
+# What the summariser is shown. Past this the oldest turns are dropped with a
+# marker: a transcript that does not fit the window it is protecting is no use.
+MAX_TRANSCRIPT_CHARS = 24_000
+
+# Settings key, in the GENERAL group. 0 turns automatic compaction off; the
+# button still works, because a person asking for it does not need a threshold.
+THRESHOLD_KEY = "compact_threshold"
+DEFAULT_THRESHOLD = 95
+
+
+def threshold(db: DBSession) -> int:
+    value = settings_store.get(db, THRESHOLD_KEY)
+    return int(value) if isinstance(value, (int, float)) else DEFAULT_THRESHOLD
+
+
+def moment(message: Message) -> datetime:
+    """A message's timestamp, always comparable.
+
+    SQLite does not store the offset, so a row loaded from disk comes back naive
+    while one still in the session's identity map keeps the tzinfo it was
+    created with. Comparing the two raises, and every comparison here is between
+    exactly those: a cutoff fetched by id against history loaded in bulk.
+    `files.sweep_orphans` already normalises for the same reason.
+    """
+    created = message.created_at
+    return created if created.tzinfo is not None else created.replace(tzinfo=UTC)
+
+
+def cutoff_message(db: DBSession, chat: Chat) -> Message | None:
+    """The message compaction reached, or None if it never has.
+
+    There is no foreign key to null this out on an upgraded database, so the
+    check is load-bearing rather than defensive: an id pointing at a message
+    that has been deleted means the boundary no longer describes anything, and
+    the chat has to read as uncompacted.
+    """
+    if not chat.compact_summary or not chat.compacted_through_id:
+        return None
+    message = db.get(Message, chat.compacted_through_id)
+    if message is None or message.chat_id != chat.id:
+        return None
+    return message
+
+
+def reset(chat: Chat) -> None:
+    """Forget that this chat was ever compacted."""
+    chat.compact_summary = ""
+    chat.compacted_through_id = None
+    chat.compacted_at = None
+
+
+def apply(chat: Chat, *, summary: str, upto: Message) -> None:
+    """Record a summary and move the boundary. Caller commits."""
+    chat.compact_summary = summary.strip()
+    chat.compacted_through_id = upto.id
+    chat.compacted_at = datetime.now(UTC)
+
+
+def split(
+    db: DBSession, chat: Chat, messages: list[Message]
+) -> tuple[list[Message], list[Message]]:
+    """(summarised, live) -- what is behind the divider, and what is not."""
+    cutoff = cutoff_message(db, chat)
+    if cutoff is None:
+        return [], list(messages)
+    boundary = moment(cutoff)
+    return (
+        [m for m in messages if moment(m) <= boundary],
+        [m for m in messages if moment(m) > boundary],
+    )
+
+
+def last_complete(db: DBSession, chat: Chat) -> Message | None:
+    """The newest finished assistant turn: where compaction should stop.
+
+    Landing on a reply rather than a question means the kept history starts on a
+    user turn, which is what every chat template expects.
+    """
+    return db.scalar(
+        select(Message)
+        .where(
+            Message.chat_id == chat.id,
+            Message.role == ROLE_ASSISTANT,
+            Message.complete.is_(True),
+        )
+        .order_by(Message.created_at.desc())
+    )
+
+
+def transcript(db: DBSession, chat: Chat, *, upto: Message) -> str:
+    """The turns to summarise, oldest first, as plain text.
+
+    Only the delta since the last compaction: the previous summary is supplied
+    separately, and the instruction asks for one record, so each summary
+    subsumes the one before it. Re-summarising the whole chat every time grows
+    quadratically and eventually exceeds the very window this protects.
+    """
+    previous = cutoff_message(db, chat)
+    query = select(Message).where(
+        Message.chat_id == chat.id,
+        Message.created_at <= upto.created_at,
+        Message.error == "",
+    )
+    if previous is not None:
+        query = query.where(Message.created_at > previous.created_at)
+
+    lines: list[str] = []
+    for message in db.scalars(query.order_by(Message.created_at)):
+        body = message.content.strip()
+        if not body:
+            continue
+        lines.append(f"{message.role}: {body}")
+
+    text = "\n\n".join(lines)
+    if len(text) > MAX_TRANSCRIPT_CHARS:
+        # Keep the most recent part: the older it is, the more likely the
+        # previous summary already covers it.
+        text = "[earlier turns omitted]\n\n" + text[-MAX_TRANSCRIPT_CHARS:]
+    return text
+
+
+def previous_summary_block(chat: Chat) -> str:
+    """The earlier summary, headed, or "" on a first compaction.
+
+    Empty is fine to pass straight through: `prompts.substitute` drops a line
+    that held a known variable and expanded to nothing, so the prompt does not
+    end up with a hole where a heading was.
+    """
+    if not chat.compact_summary.strip():
+        return ""
+    return "## Summary of even earlier turns\n\n" + chat.compact_summary.strip()
+
+
+def should_compact(db: DBSession, chat: Chat, *, pending: str = "") -> bool:
+    """Whether the next request should be summarised first.
+
+    Judged from the last reply's recorded usage plus an estimate of the new
+    turn. True prompt_tokens are only knowable after a response, so a
+    retrospective figure is the honest basis -- but on its own it is one turn
+    stale, and fifty thousand characters pasted into the composer would overflow
+    a window that measured 90% last time. The estimator covers only that delta.
+
+    Never fires when the model's context length is unknown. Acting on a number
+    nobody supplied is exactly what the 0-means-unknown rule exists to prevent.
+    """
+    limit = threshold(db)
+    if limit <= 0:
+        return False
+
+    last = last_complete(db, chat)
+    if last is None:
+        return False
+
+    usage = metrics_service.from_message(last.usage_json)
+    if usage.context_limit <= 0 or usage.context_tokens <= 0:
+        return False
+
+    projected = usage.context_tokens + tokens.estimate(pending)
+    return projected >= usage.context_limit * limit / 100

src/lembas/services/generation.py

@@ -22,9 +22,12 @@ import time
 from dataclasses import dataclass, field
 from datetime import UTC, datetime, timedelta
 
+from sqlalchemy import select
+
 from lembas.db.models import ROLE_ASSISTANT, ROLE_USER, Chat, Message, User
 from lembas.db.session import session_scope
 from lembas.services import chat as chat_service
+from lembas.services import compaction as compaction_service
 from lembas.services import metrics as metrics_service
 from lembas.services import prompts as prompts_service
 from lembas.services import tokens
@@ -95,6 +98,10 @@ class Generation:
     # woken individually: with a 100ms cadence a short poll is simpler than
     # future bookkeeping, and cannot drop a wakeup.
     version: int = 0
+    # What the reply is doing when it is not producing tokens. Shown in the
+    # streaming bubble, because a silent multi-second pause before the first
+    # token is what a hang looks like.
+    status: str = ""
     # Number of browsers currently watching. Decides whether a finished reply
     # counts as unread.
     followers: int = 0
@@ -214,6 +221,13 @@ async def _run(generation: Generation) -> None:
     title_prompt = ""
 
     try:
+        # Before the request is assembled, so build_request is called once and
+        # what goes out is the compacted conversation -- there is no second
+        # assembly path. Here rather than in post_message because that route's
+        # whole contract is to return immediately, and a three-second
+        # summarisation in front of it would break exactly that.
+        await _maybe_compact(generation)
+
         with session_scope() as db:
             chat = db.get(Chat, generation.chat_id)
             message = db.get(Message, generation.message_id)
@@ -387,6 +401,77 @@ async def _run(generation: Generation) -> None:
         generation.touch()
 
 
+async def _maybe_compact(generation: Generation) -> None:
+    """Summarise the earlier turns if the window is about to be full.
+
+    Never raises. A failed compaction logs and sends the uncompacted request,
+    which either works or fails upstream with a message that says what actually
+    happened -- refusing to answer because the summariser was unavailable would
+    be a worse trade.
+
+    The awaited call is deliberately outside any session, the same shape titling
+    uses: read everything needed, close, ask, reopen to write.
+    """
+    try:
+        with session_scope() as db:
+            chat = db.get(Chat, generation.chat_id)
+            message = db.get(Message, generation.message_id)
+            if chat is None or message is None:
+                return
+
+            pending = _pending_text(db, message)
+            if not compaction_service.should_compact(db, chat, pending=pending):
+                return
+
+            template = prompts_service.resolve(db, "task.compact")
+            upto = compaction_service.last_complete(db, chat)
+            if not template.strip() or upto is None:
+                return
+
+            endpoint, model_id = chat_service.resolve_endpoint(db, chat)
+            transcript = compaction_service.transcript(db, chat, upto=upto)
+            previous = compaction_service.previous_summary_block(chat)
+            upto_id = upto.id
+
+        generation.status = "Summarising earlier messages…"
+        generation.touch()
+
+        summary = await chat_service.summarise_for_compaction(
+            endpoint,
+            model_id,
+            transcript=transcript,
+            previous_summary=previous,
+            template=template,
+        )
+        if not summary:
+            return
+
+        with session_scope() as db:
+            chat = db.get(Chat, generation.chat_id)
+            upto = db.get(Message, upto_id)
+            if chat is None or upto is None:
+                return
+            compaction_service.apply(chat, summary=summary, upto=upto)
+            db.commit()
+            log.info("chat %s compacted automatically through %s", chat.id, upto_id)
+    except Exception:  # noqa: BLE001 - the reply matters more than the tidy-up
+        log.exception("automatic compaction failed for chat %s", generation.chat_id)
+    finally:
+        generation.status = ""
+        generation.touch()
+
+
+def _pending_text(db, message: Message) -> str:
+    """The user turn this reply is answering, for the size estimate."""
+    previous = db.scalars(
+        select(Message)
+        .where(Message.chat_id == message.chat_id, Message.created_at < message.created_at)
+        .order_by(Message.created_at.desc())
+        .limit(1)
+    ).first()
+    return previous.content if previous is not None else ""
+
+
 def _question_from(payload: dict) -> str:
     """The last thing the user said, for auto-titling."""
     for entry in reversed(payload.get("messages", [])):

src/lembas/services/prompts.py

@@ -156,6 +156,16 @@ VARIABLES: tuple[Variable, ...] = (
     ),
     Variable("question", "Question", "The first message. Chat title task only."),
     Variable("answer", "Answer", "The first reply. Chat title task only."),
+    Variable(
+        "transcript",
+        "Transcript",
+        "The turns being summarised, oldest first. Compaction task only.",
+    ),
+    Variable(
+        "previous_summary",
+        "Earlier summary",
+        "The summary from a previous compaction, if there was one. Compaction task only.",
+    ),
 )
 
 VARIABLE_NAMES = frozenset(variable.name for variable in VARIABLES)
@@ -764,6 +774,77 @@ BUILTIN: tuple[Fragment, ...] = (
             "Assistant: {{answer}}"
         ),
     ),
+    Fragment(
+        key="task.compact",
+        label="Compaction summary",
+        group=GROUP_TASKS,
+        order=410,
+        variables=("transcript", "previous_summary"),
+        hint="A separate one-message request, not part of any chat. Clear it to "
+        "turn compaction off entirely: the button says so and nothing is "
+        "summarised automatically.",
+        default=(
+            "Summarise the conversation below so it can be carried forward after the "
+            "earlier turns are dropped from your context. This is a working record, "
+            "not a report for a reader.\n"
+            "\n"
+            "Keep, under these headings and in this order:\n"
+            "\n"
+            "## What we are doing\n"
+            "The goal, and where we have got to.\n"
+            "\n"
+            "## Decisions\n"
+            "Anything settled, and why. A decision without its reason gets argued "
+            "again.\n"
+            "\n"
+            "## Facts established\n"
+            "Names, numbers, versions, file paths, URLs and identifiers, copied "
+            "exactly. Do not round them, paraphrase them or reconstruct one from "
+            "memory — if it is not in the transcript, leave it out.\n"
+            "\n"
+            "## Open threads\n"
+            "What is unfinished, and what was about to happen next.\n"
+            "\n"
+            "Leave out pleasantries, retracted ideas and anything already superseded. "
+            "Do not answer the conversation: you are recording it. Write in the "
+            "language of the conversation, and stay under 500 words.\n"
+            "\n"
+            "{{previous_summary}}\n"
+            "\n"
+            "## Transcript\n"
+            "\n"
+            "{{transcript}}"
+        ),
+    ),
+    Fragment(
+        key="task.compact_lead",
+        label="How a summary is introduced",
+        group=GROUP_TASKS,
+        order=420,
+        hint="Sits in front of the summary, in the turn that replaces the "
+        "messages no longer being sent. Without it a model reads the summary as "
+        "something the person has just typed.",
+        default=(
+            "Here is a summary of the earlier part of this conversation. Those "
+            "messages are no longer in your context. Treat this summary as an "
+            "accurate record of them and rely on it rather than on what you can no "
+            "longer see; if it does not cover something you need, say so instead of "
+            "filling the gap."
+        ),
+    ),
+    Fragment(
+        key="task.compact_ack",
+        label="The model's acknowledgement",
+        group=GROUP_TASKS,
+        order=430,
+        hint="One assistant turn after the summary, so the conversation still "
+        "alternates user, assistant, user. Several chat templates reject a "
+        "history that does not.",
+        default=(
+            "Understood. I have the summary of the earlier turns and will carry on "
+            "from there."
+        ),
+    ),
 )
 
 register_source(_builtin_source)

src/lembas/services/settings_store.py

@@ -35,6 +35,12 @@ def _general_defaults() -> dict[str, Any]:
         # Applied to every chat that has no model or chat prompt of its
         # own. See services.chat.effective_system_prompt.
         "system_prompt": "",
+        # Percentage of a model's context length at which the earlier turns are
+        # summarised automatically. 0 turns it off; the Compact button still
+        # works, because a person asking for it does not need a threshold.
+        # Never fires for a model whose context_length is 0, since that is
+        # "unknown" rather than "small". See services/compaction.py.
+        "compact_threshold": 95,
     }