通用 LLMClient.chat(messages)
任意模块传入 role: system + user 即构�?system 提示;例�?relay/settings.py 中风控为单条 user。飞鸽主路径封装�?chat_feige_* / chat_stream�?/p>
分章浏览。完整说明见 首页�?/p>
任意模块传入 为主对话注入「当前日期时间」一句(通常在组装末尾追加),便于模型对齐日历与时态;具体文案见源码格式化函数�?/p>
下列方法均有文档字符串;含「飞鸽」命名的为主对话或合�?user 轮专用封装�?strong class="aux-llm-method-list-hint">�?其它
通用
LLMClient.chat(messages)role: system + user 即构�?system 提示;例�?relay/settings.py 中风控为单条 user。飞鸽主路径封装�?chat_feige_* / chat_stream�?/p>
format_now_for_llm_system · 当前时间说明(system 尾部�?/h3>
app/routers/matches.py · _assemble_chat_system_final展开:与主对话挂载关�?· 修改备忘
【当前时间】{YYYY-MM-DD HH:mm:ss}({时区名,�?Asia/Shanghai})。今天是{date}(周×)。以今天为基准:明天=…,后天=…,大后�?…。用户说「明�?后天/大后天」时,若你要写出具体月日,必须与上列日期一致,禁止把「后天」写成其它号。若无法对齐,只复述「后天」等相对说法或向用户确认,勿编造错误公历日期。供你判断:与上文的对话间隔、约定是否到期或食言、评价用户真诚度是否随时间改善(挽回需长期稳定表现,非一时热情)�?
(实现:
app/llm_time.py · format_now_for_llm_system,时区来自环境变�?APP_TIMEZONE,默�?Asia/Shanghai。)scripts/smoke_*.py 仅用于本地验证,不在生产路径内�?/p>
LLMClient 方法速查(app/llm_client.py�?/h2>
chat_feige_* / 通用 chat 包装�?/strong>多与「单独请求」条目对应(各章常以橙黄卡片标出)�?/p>
展开:方法列�?+ 右侧可记重构/合并意向
chat / chat_stream �?通用入口(支�?system 块、Anthropic cache 前缀�?chat_feige_user_turn �?单轮用户正文
chat_feige_combined_* �?品行 JSON + 正文合并格式、工具轮
summarize_relay_similarity_angles_from_profiles / pick_best_relay_rag_record_rank
need_match_exclude_users_for_candidate_blacklists
need_match_blacklist_semantic_hit_row_indices
classify_relay_invite_reply_audience
classify_relay_short_ack_peer_or_wrapup
classify_relay_assistant_promise_wrapup_other
classify_relay_turn_mes_type
summarize_relay_content
classify_relay_invite_first_reply
classify_labeling_gate
detect_help_topic_preference_signal
extract_help_topic_setting
classify_await_help_topics_confirmation_reply
classify_peer_now_place_requirement
classify_need_segment_should_close
classify_active_user_aborts_topic
classify_labeling_confirmation_reply
classify_await_relay_confirmation_reply
pick_relevant_need_id_for_message
pick_recent_binding_candidate_for_message
classify_turn_mes_type
generate_rag_query_associations(needs_desc �?rag_query_associations �?多路 RAG 检索)
summarize_chat_session_topic(L0 切会�?�?chat_summaries�?extract_recurring_topics_from_chat_summaries_for_l1(chat_summaries �?L1�?extract_recurring_topics_from_l1_summaries_for_l2(L1 �?L2�?extract_frequent_topics_from_l1(L1 �?重要记忆�?evaluate_user_from_l1
(遗留未调用:summarize_l0_to_l1、summarize_l1_to_l2�?extract_user_needs_from_turn
extract_user_profile_from_turn
normalize_profile_salary_level_to_storage / apply_salary_tool_normalization_to_profile_payload
evaluate_user_from_l1 / evaluate_user_from_messages
infer_user_merchant_preference_score
infer_merchant_communication_attitude_snippet_only
infer_merchant_communication_preference_and_attitude_snippet
_structure_merchant_kb_rag_summary_lines / structure_merchant_kb_plaintext_for_rag / structure_merchant_kb_both_formats_for_rag
for_merchant_kb_structure(类方法,切换商�?DeepSeek 客户端)
文档目录�?code>docs/llm-call-catalog/