feat(turbo): add direct LLM chat mode with runtime toggle

Add 极速 mode that streams via /api/direct-chat without the lead agent stack,
plus VITE_TURBO_MODE_VISIBLE in runtime-config.js (default off). Also disable
global memory, tighten Qwen/vLLM thinking control for flash mode, and force
thinking off on the direct-chat endpoint.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
董宏凯andCursor committed 2026-09-09 21:58:36 +08:00
1 parent 4f7f07504a
commit d765afe2ac
23 files changed
+1295 -394

No files matched your search

@@ -16,7 +16,7 @@
============
``get_effective_memory_config(user_id)`` 按优先级合并 (高 -> 低):
1. per-user 覆盖 数据库 ``user_memory_configs`` 表 (经进程内 cache 读取)
(有 per-user 记录时自动强制 enabled=True)
(仅白名单字段,不能覆盖 ``enabled``)
2. config.yaml ``memory:`` 段 (即 ``get_memory_config()``)
3. Pydantic 默认值
@@ -254,13 +254,12 @@ def get_effective_memory_config(user_id: str) -> MemoryConfig:
"""返回某用户的生效记忆配置 (同步,从进程内 cache 读取 per-user 覆盖)。
优先级 (高 -> 低):per-user DB 覆盖 (经 cache) > config.yaml > 默认值。
有 per-user 记录时自动强制 enabled=True。
``enabled`` 始终以 config.yaml 为准,用户覆盖不能把它重新打开。
"""
base = _memory_config.model_dump()
recall_budget = _recall_budget_cache.get(user_id)
if recall_budget is not None:
base.setdefault("hindsight", {})["recall_budget"] = recall_budget
base["enabled"] = True
try:
return MemoryConfig(**base)
except Exception as exc: # noqa: BLE001 - 覆盖产生非法配置时降级