feat(avatar): understand BOXIM image messages

This commit is contained in:
stefanfeng
2026-09-01 15:35:41 +08:00
parent 3d999f9472
commit 0fc43908ae
8 changed files with 523 additions and 13 deletions
@@ -3,6 +3,7 @@
import asyncio
import hashlib
import logging
import os
import re
import secrets
import time
@@ -13,12 +14,19 @@ from sqlalchemy.orm import Session
from models import (
Avatar,
ChatAttachment,
TakeoverCursor,
TakeoverMessage,
TakeoverReplyTask,
User,
)
from services.boxim_client import BoxIMClient, BoxIMError
from services.boxim_image_service import (
BoxIMImageError,
download_boxim_image,
parse_boxim_image_url,
)
from services.vision_service import ImageValidationError
logger = logging.getLogger(__name__)
@@ -37,6 +45,10 @@ HUMAN_PAUSE_SECONDS = 600
RATE_LIMIT_WINDOW_SECONDS = 300
RATE_LIMIT_MAX_REPLIES = 5
AVATAR_LOCAL_ID_PREFIX = "880"
BOXIM_TEXT_MESSAGE_TYPE = 0
BOXIM_IMAGE_MESSAGE_TYPE = 1
BOXIM_IMAGE_PROMPT = "请看看这张图片。"
BOXIM_IMAGE_UNAVAILABLE_REPLY = "这张图片我暂时没看清,麻烦重新发送一张清晰的原图。"
def _utcnow() -> datetime:
@@ -107,6 +119,14 @@ def _configured_reply_delay(avatar: Avatar, fallback: int | None = None) -> int:
return delay
def _event_prompt(event: TakeoverMessage) -> str:
if event.message_type == BOXIM_TEXT_MESSAGE_TYPE:
return event.content.strip()
if event.message_type == BOXIM_IMAGE_MESSAGE_TYPE:
return BOXIM_IMAGE_PROMPT
return ""
class TakeoverService:
"""Poll BOXIM, honor the owner grace period, then generate and send one reply."""
@@ -498,8 +518,27 @@ class TakeoverService:
if not is_avatar:
self._cancel_conversation(db, avatar.owner_id, peer_id, "owner_replied")
return
if not schedule_reply or event.message_type != 0 or not event.content.strip():
if not schedule_reply or event.message_type not in {
BOXIM_TEXT_MESSAGE_TYPE,
BOXIM_IMAGE_MESSAGE_TYPE,
}:
return
if event.message_type == BOXIM_TEXT_MESSAGE_TYPE and not event.content.strip():
return
if event.message_type == BOXIM_IMAGE_MESSAGE_TYPE:
try:
parse_boxim_image_url(
event.content,
base_url=getattr(self.boxim, "im_base_url", ""),
)
except BoxIMImageError as exc:
logger.warning(
"Ignored invalid BOXIM image message %s for avatar %s: %s",
message_id,
avatar.id,
exc,
)
return
if (now - send_time).total_seconds() > self.max_message_age_seconds:
logger.info(
"Ignored stale BOXIM message %s for avatar %s (age=%ss)",
@@ -606,7 +645,7 @@ class TakeoverService:
task.status = "cancelled"
task.cancel_reason = "newer_incoming_message"
task.locked_at = None
prompt_parts.append(event.content.strip())
prompt_parts.append(_event_prompt(event))
source_ids.append(event.boxim_message_id)
prompt = "\n".join(part for part in prompt_parts if part).strip()[-MAX_PROMPT_LENGTH:]
due_at = max(
@@ -665,6 +704,50 @@ class TakeoverService:
results = await asyncio.gather(*(generate(task_id) for task_id in task_ids))
return sum(bool(result) for result in results)
def _takeover_image_attachment(
self,
db: Session,
avatar: Avatar,
event: TakeoverMessage,
) -> ChatAttachment:
now = self.now()
if event.attachment_id:
cached = db.get(ChatAttachment, event.attachment_id)
if cached and cached.status == "ready" and cached.expires_at > now:
cached.used_at = now
db.commit()
return cached
downloaded = download_boxim_image(
event.content,
base_url=getattr(
self.boxim,
"im_base_url",
os.getenv("BOXIM_API_BASE_URL", "https://im.99hui.com/api"),
),
)
from routers.chat import _analyze_image_bytes
attachment = _analyze_image_bytes(
db,
avatar,
downloaded.content,
filename=downloaded.filename,
mime_type=downloaded.mime_type,
uploader_kind="boxim",
)
event.attachment_id = attachment.id
attachment.used_at = now
db.commit()
logger.info(
"BOXIM image analyzed message=%s attachment=%s avatar=%s category=%s",
event.boxim_message_id,
attachment.id,
avatar.id,
attachment.category,
)
return attachment
def _generate_reply(self, task_id: str) -> bool:
db = self.session_factory()
try:
@@ -694,9 +777,36 @@ class TakeoverService:
.limit(30)
.all()
)
source_events = {
event.boxim_message_id: event
for event in events
if event.boxim_message_id in excluded_ids
}
image_attachments = []
image_failed = False
for message_id in (task.source_message_ids or [])[-3:]:
event = source_events.get(message_id)
if not event or event.message_type != BOXIM_IMAGE_MESSAGE_TYPE:
continue
try:
image_attachments.append(
self._takeover_image_attachment(db, avatar, event)
)
except (BoxIMImageError, ImageValidationError) as exc:
image_failed = True
logger.warning(
"BOXIM image unavailable message=%s avatar=%s: %s",
event.boxim_message_id,
avatar.id,
exc,
)
history = []
for event in reversed(events):
if event.boxim_message_id in excluded_ids or not event.content.strip():
if (
event.boxim_message_id in excluded_ids
or event.message_type != BOXIM_TEXT_MESSAGE_TYPE
or not event.content.strip()
):
continue
if event.direction == "incoming" and event.is_avatar:
continue
@@ -708,10 +818,25 @@ class TakeoverService:
)
history = history[-10:]
from routers.chat import _resolve_reply
from routers.chat import _attachment_contexts, _resolve_reply
result = _resolve_reply(db, avatar, task.prompt, history, usage_source="takeover")
answer = _plain_text_reply(result.get("answer", ""))
image_contexts = _attachment_contexts(image_attachments)
has_source_text = any(
event.message_type == BOXIM_TEXT_MESSAGE_TYPE and event.content.strip()
for event in source_events.values()
)
if image_failed and not image_contexts and not has_source_text:
answer = BOXIM_IMAGE_UNAVAILABLE_REPLY
else:
result = _resolve_reply(
db,
avatar,
task.prompt,
history,
usage_source="takeover",
image_contexts=image_contexts,
)
answer = _plain_text_reply(result.get("answer", ""))
db.refresh(task)
if task.status != "generating":
return False