* fix(server): drop the Chinese seed from the chat title prompt (MUL-5689)
chatTitleSystemPrompt carried literal Chinese in two places: a rule that
spelled out "Chinese input → Chinese title", and a formatting example
listing "标题：" as a prefix not to use. It is the same defect as the
quick-actions label rule — a prompt that names a language, even only as a
formatting example, reads as permission to answer in it.
Chat titles are generated from the user's opening message alone, so they
have neither the ALREADY SUGGESTED feedback loop nor an agent reply to be
pulled by. Removing the seed is the whole fix; the language rule stays,
now stated without naming a language.
Nothing is lost by dropping the "标题：" example: chatTitleLabelPrefixes
already strips 标题/题目/主题 (and the English forms) from the model's
output, which is where that guarantee actually lives.
Co-authored-by: multica-agent <github@multica.ai>
* fix(server): correct two inaccuracies in the quick-actions language rule
Both from review on #6345, non-blocking there and deferred to keep that
merge unblocked.
"these instructions" was self-referential: the LANGUAGE RULE is itself an
instruction, so a model reading "ignore these instructions when choosing
the language" could read the rule as disowning itself. It means the
system prompt, so it now says so.
The system prompt claimed the user message "ends with" the LANGUAGE RULE.
It does not — the task line follows it. Reworded to "contains a LANGUAGE
RULE line near the end", which is what the renderer actually produces.
The rule's position is unchanged: still after the conversation and the
replayed labels, which is the part that matters.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Bohan-J <bohan@devv.ai>
Co-authored-by: multica-agent <github@multica.ai>
* feat(chat): LLM-generated chat session titles with silent fallback (MUL-4295)
Generate a concise, language-matched title for a chat session after the
first user message, replacing the raw first-message-derived title. The
work is best-effort and fully non-blocking:
- Triggered on the first user message in SendChatMessage (detected via
ChatSessionHasUserMessage before insert), run in a detached goroutine
so it never delays the send or first response.
- Reuses pkg/llm GenerateText on the configured default model
(MULTICA_LLM_DEFAULT_MODEL, else gpt-4o-mini); no model from the client.
- Self-hosted with no LLM key (h.LLM.Enabled()==false): silent no-op,
the original title stands. Same on timeout / upstream error.
- CAS write (UpdateChatSessionTitleIfCurrent) so a manual rename during
generation is never clobbered and titling runs at most once.
- Pushes chat:session_updated so the frontend refreshes in place.
- sanitizeChatTitle strips quotes/brackets, 'Title:'/'标题：' prefixes,
trailing punctuation, and caps at chatSessionTitleMaxLen.
Tests cover all six cases: configured→semantic title, disabled→fallback,
upstream error→fallback, manual rename→no clobber, empty output→fallback,
idempotent second run, plus sanitize rules and the realtime push.
Co-authored-by: multica-agent <github@multica.ai>
* fix(chat): panic-contain title goroutine + loop sanitizer to a fixed point (MUL-4295)
Address PR #5141 review (张大彪 / multica-eve, Phase B):
1. The detached title-generation goroutine now has a defer recover() at the
top of its body. It runs outside chi's Recoverer, so an unhandled panic
in GenerateText / sanitize / the DB write / publish would crash the
server process. Best-effort path: log and keep the original title.
2. sanitizeChatTitle now alternates prefix-stripping and wrapper-stripping
in a loop until the string is stable, so a forbidden label hidden inside
a wrapper ("Title: Fix login", 「标题：修复登录问题」) is fully cleaned
regardless of nesting order. Added both cases to the sanitize test table.
Co-authored-by: multica-agent <github@multica.ai>
* fix(chat): fold trailing-punctuation trim into sanitizer fixed-point loop (MUL-4295)
Address PR #5141 follow-up review: the trailing-punctuation trim ran once
AFTER the prefix/wrapper loop, so a trailing '.' / '。' left the closing
wrapper unrecognized and the forbidden prefix untouched for inputs like
"Title: Fix login". and 「标题：修复登录问题」。. Trailing trim now runs inside the
same loop, so removing the trailing punctuation re-exposes the wrapper (and
the prefix it hid) on the next pass. Added both cases to TestSanitizeChatTitle.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
