Author SHA1 Message Date
arys 1dcfaf46cf feat: no-answer retry, agent status machine, escalation events/timeline (ТЗ §13-15,22,25-27,37, AC-08,16-19)
Phase 2 of the L1->L2 routing engine (Phase 1: MR!4).

- ami_loop() now also captures native AMI DialEnd/Hangup frames (not
  only UserEvent), needed to detect that an escalated agent did not
  answer. No dialplan change required - Redirect already routes the
  client channel into an existing Dial()-based transfer context, so
  Asterisk emits these events on its own; the listener just wasn't
  reading them before.
- retry_escalation_no_answer(): on NOANSWER/BUSY/CANCEL/CHANUNAVAIL/
  CONGESTION, releases the non-answering agent, excludes it, and
  reserves+redirects to the next available agent via the routing
  engine's existing exclude_agent_ids support. Exhausted pool marks
  the escalation failed and leaves the call with the AI instead of
  dropping the client (ТЗ §32).
- Agent status now actually moves through
  RESERVED -> RINGING -> TALKING -> AFTER_CALL_WORK -> AVAILABLE
  instead of staying stuck on RESERVED for the whole call; a new
  acw_sweep_loop background thread (same pattern as the existing
  failed_retry_loop) times out AFTER_CALL_WORK back to AVAILABLE.
- escalations gains attempt_count/real_agent_id/attempted_agent_ids_json
  (migration 0033); fixes a latent bug where assigned_agent_id stored
  the SIP extension instead of the real agent_id despite routing-service
  already returning it in RoutingAgentReserveOut.
- Every transition now records an interaction timeline entry and
  publishes the ТЗ §25 event catalog (AgentReserved/AgentRinging/
  AgentNoAnswer/AgentConnected/TransferCompleted/TransferFailed)
  through the existing emit_voice_event/EventOutboxRow idempotent path.

Not in this MR (see plan): SLA config, Callback, L3 (needs real
technical agents from the business), metrics.
2026-08-30 14:19:23 +05:00
didar 010a8dcab6 fix: ensure emotive ack rotation and prebaked ack caching apply outside v2 queue eligibility
deploy / deploy (push) Successful in 30s
2026-08-30 12:49:26 +05:00
Codex 13ba8f8b56 merge: bring aimaq host-override files back into git (fix/aimaq-remove-host-override)
deploy / deploy (push) Successful in 32s
2026-08-30 07:46:21 +00:00
arystanbek 9d918af12a fix: bring aimaq host-override files (voice.py, persona env) back into git
Both services/ai_orchestrator_service/voice.py and the aimaq persona/DOMAIN
SCOPE prompt were bind-mounted straight from the host on the aimaq stack,
bypassing git and CI entirely since they were first hand-edited in prod.

voice.py: merged the host's live business logic (gas/aimaq domain keyword
list, off-domain Kazakh/Russian replies, disabled re-correction of an already
obtained name) with the timeout_seconds fix from 09bcf74 that never reached
aimaq because the bind mount blocked it.

aimaq.env.production: replaced the AI_OPERATOR_* env values (which were never
interpolated -- {agent_name}/{company_name} would have been read literally)
with the final resolved Zhanna/Kazakgaz Aimaq text including the DOMAIN SCOPE
clause, matching what was actually live on the host.
2026-08-30 07:45:35 +00:00
arystanbek ff85fa27fb Merge pull request 'feat: aimaq voice AI persona Zhanna + language-choice greeting' (#7) from feature/aimaq-zhanna-persona-language-choice into main
deploy / deploy (push) Successful in 30s
2026-08-30 07:22:16 +00:00
arys 82c89eff5a feat: give aimaq voice AI a Qazaqgaz Aimaq persona (Zhanna) and language-choice greeting
Parametrize ai_operator_default_config() via AI_OPERATOR_* env vars,
falling back to the existing hardcoded defaults so the other two stacks
(call-center, sales-call-center) that share this code are unaffected.

Set aimaq-only overrides matching the client-provided script (Скрипт
ии-оператора КГА.docx): agent renamed to Жанна, company to Казакгаз
Аймак, the voice greeting now asks the caller whether Russian or
Kazakh is more convenient before anything else, and the base system
prompt instructs the model to commit to whichever language the caller
picks for the rest of the call, ask how to address them, and confirm
there's nothing else before saying goodbye.
2026-08-30 12:22:00 +05:00
didar 36c2acb011 test: add test for emotive ack rotation not gated on v2 queue eligibility
deploy / deploy (push) Successful in 31s
2026-08-30 12:03:37 +05:00
didar fe9b3f3a80 feat: enhance voice reply logic to prevent duplicate name addressing and improve greeting handling
deploy / deploy (push) Successful in 32s
2026-08-30 11:53:56 +05:00
arystanbek 0a829d26c7 Merge pull request 'fix: TTS ack-bank stale voice cache + website/English pronunciation' (#6) from fix/voice-tts-ack-cache-and-pronunciation into main
deploy / deploy (push) Successful in 30s
2026-08-30 06:39:32 +00:00
arys fe8598e09f fix: TTS ack-bank served stale voice after voice-config changes; add website/English pronunciation rules
RuntimeConfiguredTTSProvider (the live 'dynamic' TTS provider, resolves
voice from DB-backed config) never overrode cache_fingerprint(), so it
fell back to the base class's empty string. PrebakedAckBank keys its
on-disk cache on that fingerprint, so short filler phrases like
'Секунду' kept serving audio baked with the previous ElevenLabs voice
even after a voice change, while full LLM replies (cached inside the
resolved provider itself, keyed on its own voice id) already used the
new voice — explaining why callers heard two different voices in the
same call. Fix: delegate cache_fingerprint() to the resolved provider.

Also extend the voice delivery_hint so the model transliterates website
addresses and English words/abbreviations into spoken Cyrillic instead
of leaving raw Latin text for the TTS engine to mangle (egov.kz was
coming out as 'эговкз').
2026-08-30 11:39:14 +05:00
didar 9e4c47eddd feat: enhance AudioSocketMediaRuntime to skip filler acks for closing intents and throttle repeated filler acks
deploy / deploy (push) Successful in 30s
2026-08-30 11:30:55 +05:00
arystanbek a013ac95e8 Merge pull request 'fix: voice AI spells out numbers/dates for TTS' (#5) from fix/voice-number-pronunciation into main
deploy / deploy (push) Successful in 31s
2026-08-30 06:22:05 +00:00
29 changed files with 1391 additions and 75 deletions
+13
View File
@@ -0,0 +1,13 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"title": "AgentConnected",
"type": "object",
"required": ["event", "call_id", "escalation_id", "agent_id", "created_at"],
"properties": {
"event": { "const": "AgentConnected" },
"call_id": { "type": "string" },
"escalation_id": { "type": "string" },
"agent_id": { "type": "string" },
"created_at": { "type": "string", "format": "date-time" }
}
}
+15
View File
@@ -0,0 +1,15 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"title": "AgentNoAnswer",
"type": "object",
"required": ["event", "call_id", "escalation_id", "agent_id", "dial_outcome", "created_at"],
"properties": {
"event": { "const": "AgentNoAnswer" },
"call_id": { "type": "string" },
"escalation_id": { "type": "string" },
"agent_id": { "type": ["string", "null"] },
"dial_outcome": { "type": "string" },
"attempt_count": { "type": "integer" },
"created_at": { "type": "string", "format": "date-time" }
}
}
+18
View File
@@ -0,0 +1,18 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"title": "AgentReserved",
"type": "object",
"required": ["event", "call_id", "escalation_id", "agent_id", "created_at"],
"properties": {
"event": { "const": "AgentReserved" },
"call_id": { "type": "string" },
"escalation_id": { "type": "string" },
"agent_id": { "type": "string" },
"tenant_id": { "type": ["string", "null"] },
"from_level": { "type": "string" },
"to_level": { "type": "string" },
"reason_code": { "type": "string" },
"attempt_count": { "type": "integer" },
"created_at": { "type": "string", "format": "date-time" }
}
}
+13
View File
@@ -0,0 +1,13 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"title": "AgentRinging",
"type": "object",
"required": ["event", "call_id", "escalation_id", "agent_id", "created_at"],
"properties": {
"event": { "const": "AgentRinging" },
"call_id": { "type": "string" },
"escalation_id": { "type": "string" },
"agent_id": { "type": "string" },
"created_at": { "type": "string", "format": "date-time" }
}
}
+13
View File
@@ -0,0 +1,13 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"title": "TransferCompleted",
"type": "object",
"required": ["event", "call_id", "escalation_id", "agent_id", "created_at"],
"properties": {
"event": { "const": "TransferCompleted" },
"call_id": { "type": "string" },
"escalation_id": { "type": "string" },
"agent_id": { "type": "string" },
"created_at": { "type": "string", "format": "date-time" }
}
}
+15
View File
@@ -0,0 +1,15 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"title": "TransferFailed",
"type": "object",
"required": ["event", "call_id", "created_at"],
"properties": {
"event": { "const": "TransferFailed" },
"call_id": { "type": "string" },
"escalation_id": { "type": ["string", "null"] },
"agent_id": { "type": ["string", "null"] },
"reason": { "type": ["string", "null"] },
"error": { "type": ["string", "null"] },
"created_at": { "type": "string", "format": "date-time" }
}
}
+7
View File
@@ -185,3 +185,10 @@ EVENT_BUS_EXCHANGE=mvpcc.domain.events
ERROR_TELEGRAM_ALERTS=0
ERROR_TELEGRAM_ALERT_BOT_TOKEN=8734026216:AAEK-xdgVR5al6rfKp_DIb-HSJZXGUG2Tvg
ERROR_TELEGRAM_ALERT_CHANNEL_ID=t.me/konturaitelecom
AI_OPERATOR_AGENT_NAME=Жанна
AI_OPERATOR_COMPANY_NAME=Казакгаз Аймак
AI_OPERATOR_VOICE_GREETING_RU=Добрый день! Меня зовут Жанна, я ИИ-оператор контакт-центра Казакгаз Аймак. Вам удобнее получить консультацию на казахском или на русском языке?
AI_OPERATOR_VOICE_GREETING_KZ=Қайырлы күн! Менің атым Жанна, мен «Қазақгаз Аймақ» байланыс орталығында жасанды интеллект операторымын. Кеңесті қазақ немесе орыс тілінде алғыңыз келе ме?
AI_OPERATOR_IDENTITY_REPLY_RU=Я Жанна, ИИ-оператор контакт-центра Казакгаз Аймак. Чем могу помочь?
AI_OPERATOR_IDENTITY_REPLY_KZ=Мен Жаннамын, «Қазақгаз Аймақ» байланыс орталығының жасанды интеллект операторымын. Қалай көмектесе аламын?
AI_OPERATOR_BASE_SYSTEM_PROMPT=Ты Жанна, единый ИИ-оператор контакт-центра Казакгаз Аймак для звонков, Telegram и других каналов. Всегда сохраняй одну и ту же личность: тебя зовут Жанна. Если клиент спрашивает, кто ты или как тебя зовут, отвечай, что ты Жанна. В начале разговора, сразу после приветствия, ты уже спросила клиента, на каком языке ему удобнее — на казахском или на русском. Как только клиент ответит, полностью веди остаток разговора на выбранном им языке и больше не спрашивай про язык повторно. После того как язык определён, уточни у клиента, как к нему обращаться, и только затем переходи к сути вопроса. Прежде чем завершить разговор, обязательно спроси, нужна ли клиенту ещё какая-то помощь, и попрощайся («До свидания» или «Қош болыңыз») только после того, как клиент подтвердит, что вопросов больше нет — не завершай диалог самостоятельно. Отвечай естественно, кратко и по делу. Когда говоришь о себе, используй женский род: могла, смогла, сделала, готова, проверила, нашла. Телефонные номера читай по цифрам. Не используй Markdown, URL или таблицы. DOMAIN SCOPE: ты отвечаешь ТОЛЬКО на вопросы, связанные с услугами газоснабжения: оплата газа, счётчики, ваучеры, технические условия, отключение газа, мобильное приложение, квитанции, передача показаний, контакты филиалов, безопасность газа. Если вопрос НЕ связан с газом (погода, спорт, политика, другие компании, личные советы, кулинария, рецепты, анекдоты, новости, фильмы), вежливо откажи БЕЗ поиска в базе знаний и БЕЗ попытки ответить: RU «Я могу ответить только на вопросы, связанные с услугами газоснабжения. Пожалуйста, задайте вопрос по теме газа.» KK «Мен тек газмен қамтамасыз ету қызметтеріне байланысты сұрақтарға жауап беремін. Газ тақырыбында сұрақ қойыңызшы.» Не пытайся ответить на вопрос не по теме, даже если знаешь ответ.
@@ -0,0 +1,4 @@
ALTER TABLE escalations ADD COLUMN IF NOT EXISTS attempt_count INTEGER NOT NULL DEFAULT 0;
ALTER TABLE escalations ADD COLUMN IF NOT EXISTS real_agent_id TEXT;
ALTER TABLE escalations ADD COLUMN IF NOT EXISTS attempted_agent_ids_json TEXT NOT NULL DEFAULT '[]';
CREATE INDEX IF NOT EXISTS ix_escalations_real_agent_id ON escalations(real_agent_id);
@@ -0,0 +1,4 @@
ALTER TABLE escalations ADD COLUMN attempt_count INTEGER NOT NULL DEFAULT 0;
ALTER TABLE escalations ADD COLUMN real_agent_id TEXT;
ALTER TABLE escalations ADD COLUMN attempted_agent_ids_json TEXT NOT NULL DEFAULT '[]';
CREATE INDEX IF NOT EXISTS ix_escalations_real_agent_id ON escalations(real_agent_id);
@@ -143,7 +143,12 @@ def operator_system_prompt(*, language: str, channel_label: str, is_voice: bool,
"Short hotline or service numbers (e.g. 1414, 109) must be spelled out the way people say them as a code, "
"grouped and read naturally (\"1414\" as \"четырнадцать четырнадцать\", not \"тысяча четыреста четырнадцать\"). "
"Calendar dates must use the correct spoken grammatical case (\"25 числа\" as \"двадцать пятого числа\", "
"not \"двадцать пять число\"; \"14 марта\" as \"четырнадцатого марта\")."
"not \"двадцать пять число\"; \"14 марта\" as \"четырнадцатого марта\"). "
"Never leave a website address, domain, or English word/abbreviation in raw Latin script — the TTS engine "
"slurs it into gibberish (e.g. \"egov.kz\" comes out as \"эговкз\"). Transliterate it into how a person "
"actually pronounces it aloud, with an explicit pause word for punctuation: write \"egov.kz\" as "
"\"игов точка кэ-зэт\", write \".kz\"/\".com\" as \"точка кэ-зэт\"/\"точка ком\", spell out an acronym or "
"English word phonetically in Cyrillic (\"IT\" as \"ай-ти\", \"email\" as \"имейл\")."
if is_voice
else "The reply should read like a concise message from a live first-line operator."
)
+114 -20
View File
@@ -603,20 +603,42 @@ def _voice_has_name_correction(text: str | None) -> bool:
return any(marker in normalized for marker in correction_markers)
def _voice_reply_with_name(language: str, reply_text: str, name: str | None) -> str:
def _voice_reply_already_names_customer(reply_text: str, normalized_short: str) -> bool:
tokens = _voice_text_key(reply_text).split(" ")
if normalized_short in tokens:
return True
stem_len = max(len(normalized_short) - 2, 3)
stem = normalized_short[:stem_len]
return any(len(token) >= stem_len and token.startswith(stem) for token in tokens)
def _voice_greeting_word(language: str) -> str:
if str(language or "").strip().lower() == "kz":
return "Сәлеметсіз бе"
return "Здравствуйте"
def _voice_reply_with_name(
language: str,
reply_text: str,
name: str | None,
*,
greet: bool = False,
) -> str:
short_name = _voice_short_name(name)
if not short_name:
return reply_text
prefix = _voice_disclosure_prefix(language)
normalized_short = _voice_text_key(short_name)
lead = f"{_voice_greeting_word(language)}, {short_name}" if greet else short_name
if reply_text.startswith(prefix):
rest = reply_text[len(prefix) :].lstrip()
if normalized_short in _voice_text_key(rest).split(" "):
if _voice_reply_already_names_customer(rest, normalized_short):
return reply_text
return f"{prefix}{short_name}, {rest}"
if normalized_short in _voice_text_key(reply_text).split(" "):
return f"{prefix}{lead}, {rest}"
if _voice_reply_already_names_customer(reply_text, normalized_short):
return reply_text
return f"{short_name}, {reply_text}"
return f"{lead}, {reply_text}"
def _voice_name_metadata(
@@ -746,13 +768,7 @@ def _voice_downstream_name_update(
action = "provide" if not name_value else "confirm"
candidate = candidate or name_value
elif status == "name_obtained":
if (
candidate
and candidate_key
and candidate_key != current_key
and _voice_has_name_correction(transcript_text)
):
action = "correct"
pass
else:
if explicit_candidate:
action = "provide"
@@ -1039,7 +1055,6 @@ def _voice_has_service_topic(text: str | None) -> bool:
"ошибк",
"проблем",
"сбой",
"интернет",
"связь",
"оператор",
"менеджер",
@@ -1047,6 +1062,40 @@ def _voice_has_service_topic(text: str | None) -> bool:
"компан",
"подключ",
"доставк",
"газ",
"счётчик",
"счетчик",
"ваучер",
"отключ",
"квитанц",
"показан",
"приложен",
"безопасн",
"техническ",
"плит",
"котл",
"труб",
"утечк",
"запах",
"абонент",
"договор",
"поверк",
"монтаж",
"счёт",
"счет",
"долг",
"задолжен",
"перерасчёт",
"перерасчет",
"регион",
"аимак",
"aimaq",
"qazaqgaz",
"казахгаз",
"есептегіш",
"төлеу",
"өтінім",
"шарт",
)
return any(marker in normalized for marker in service_markers)
@@ -1080,6 +1129,37 @@ def _voice_is_off_domain_request(text: str | None) -> bool:
"java",
"javascript",
"погод",
"плов",
"приготов",
"кулинар",
"блюдо",
"фильм",
"кино",
"спорт",
"футбол",
"хоккей",
"теннис",
"баскетбол",
"президент",
"политик",
"выборы",
"правительств",
"курс валют",
"криптовалют",
"биткоин",
"песн",
"музык",
"танц",
"шутк",
"загадк",
"сериал",
"книга",
"стихотвор",
"чемпионат",
"ауа райы",
"аспаздық",
"кітап",
"ән айт",
)
if any(marker in normalized for marker in broad_markers):
return True
@@ -1089,7 +1169,14 @@ def _voice_is_off_domain_request(text: str | None) -> bool:
"что такое",
"объясни",
"расскажи про",
"почему",
"посоветуй",
"как приготовить",
"какая погода",
"кто президент",
"кто выиграл",
"какая команда",
"ауа райы қалай",
"әнді айт",
)
return any(normalized.startswith(prefix) for prefix in broad_openers)
@@ -1097,13 +1184,12 @@ def _voice_is_off_domain_request(text: str | None) -> bool:
def _voice_off_domain_reply(language: str) -> tuple[str, str]:
if language == "kz":
return (
"Men kompaniyamyzdyn qyzmetteri men otinishteri boiynsha komek bere alamyn. "
"Eger suraq bizdin qyzmetke qatysty bolsa, qysqasha naqtylaңыз. Qalasaңыз, operatorga qosamyn.",
"AI qongyraudyn taqyrybyn kompaniya qyzmetteri sheginde naqtylaudy usyndy.",
"Кешіріңіз, мен тек газ қызметтері бойынша көмектесемін: төлеу, есептегіш, ваучер. Қалай көмектесе аламын?",
"AI off-topic сұрауды газ қызметтері тақырыбына шектеді.",
)
return (
"Я помогу по вопросам наших услуг и обращений. Если вопрос связан с нашей компанией, скажите коротко, что именно нужно. Если хотите, сразу соединю с оператором.",
"AI мягко вернул разговор к вопросам компании и предложил перевод на оператора.",
"Извините, я помогаю только по вопросам газа: оплата, счётчики, ваучеры. Чем могу помочь?",
"AI отклонил off-topic вопрос и ограничил тему газоснабжением.",
)
@@ -1613,6 +1699,12 @@ def _voice_llm_prompt_messages(
is_voice=True,
config=operator_config,
)
system_prompt += (
" Do not open `reply_text` with a greeting or by addressing the customer by name "
"(e.g. do not write 'Здравствуйте, <имя>' or start with '<имя>,'). The system inserts "
"the customer's name into the spoken reply separately, so naming them yourself would "
"make it get said twice."
)
if name_status in ("name_not_obtained", "name_followup_required"):
system_prompt += (
" The user's name is not yet obtained. If the user explicitly provided their name in this turn, "
@@ -1810,7 +1902,7 @@ def _voice_decision(
"latency_ms": 1,
}
if not kb_results and _voice_is_off_domain_request(normalized):
if _voice_is_off_domain_request(normalized):
reply_text, summary_text = _voice_off_domain_reply(language)
decision = {
"language": language,
@@ -2500,10 +2592,12 @@ def turn_voice_session(session_id: str, payload: VoiceAITurnIn) -> VoiceAITurnDe
decision_metadata.update(decision.get("metadata") or {})
suppress_name_prefix = bool(request_metadata.get("suppress_name_prefix")) if isinstance(request_metadata, dict) else False
if effective_name_status == "name_obtained" and effective_name_value and not early_plan_only and not suppress_name_prefix:
just_learned_name = current_name_status != "name_obtained"
decision["reply_text"] = _voice_reply_with_name(
decision["language"],
decision["reply_text"],
effective_name_value,
greet=just_learned_name,
)
elif inline_name_followup and not decision["needs_handoff"] and not early_plan_only:
inline_followup = _voice_inline_name_followup(decision["language"], config)
+9 -2
View File
@@ -822,8 +822,15 @@ def _media_registration_from_row(row: VoiceAISessionRow, *, queue_code: str | No
voice_v2_partial_asr=bool(voice_v2_for_session and _voice_v2_partial_asr_enabled()),
voice_v2_duplex=bool(voice_v2_for_session and _voice_v2_duplex_enabled()),
voice_v2_streaming_asr_backend=streaming_backend,
voice_v2_prebaked_ack=bool(voice_v2_for_session and _voice_v2_prebaked_ack_enabled()),
voice_v2_emotive_ack=bool(voice_v2_for_session and _voice_v2_emotive_ack_enabled()),
# voice_v2_prebaked_ack and voice_v2_emotive_ack are deliberately NOT gated on
# voice_v2_for_session: filler-ack synthesis/caching and phrase-variant rotation
# only need a text pool + the ack bank, not the v2 duplex/partial-ASR pipeline.
# Without this, calls outside the v2 queue allowlist would synthesize every
# filler live via ElevenLabs before it could play — adding real TTS round-trip
# time to the one phrase whose whole job is to hide that latency — and would
# always get the single fixed "Секунду." fallback string instead of rotating.
voice_v2_prebaked_ack=bool(_voice_v2_prebaked_ack_enabled()),
voice_v2_emotive_ack=bool(_voice_v2_emotive_ack_enabled()),
voice_v2_emotive_ack_ru_only=bool(_voice_v2_emotive_ack_ru_only()),
)
@@ -217,6 +217,10 @@ class AudioSocketMediaRuntime:
self._immediate_ack_min_ms = 700
self._v2_ack_post_gap_seconds = 0.10
self._v1_ack_wait_seconds = 0.6
# Skip a would-be filler if the previous one finished too recently, so rapid
# back-and-forth turns (e.g. a caller spelling out a phone number field by
# field) don't get a filler read before every single fragment.
self._ack_min_repeat_gap_seconds = 2.5
self._partial_poll_interval_seconds = 0.20
# Small startup cushion for streamed TTS playback: absorb ElevenLabs
# network delivery jitter before we start pacing frames out to the
@@ -314,10 +318,32 @@ class AudioSocketMediaRuntime:
return True
return normalized in cls._FINAL_LOW_SIGNAL_PHRASES
_CLOSING_INTENT_PHRASES = (
"до свидания",
"всего доброго",
"хорошего дня",
"хорошего вечера",
"прощайте",
"созвонимся",
"это все спасибо",
"это всё спасибо",
"у меня все спасибо",
"у меня всё спасибо",
"больше вопросов нет",
"вопросов больше нет",
"спасибо за помощь",
"спасибо большое до свидания",
"сау болыңыз",
"келесіге дейін",
"рахмет көп",
)
def _detect_early_intent(self, text: str) -> str:
normalized = self._normalize_intent_text(text)
if not normalized:
return "unknown"
if any(token in normalized for token in self._CLOSING_INTENT_PHRASES):
return "closing"
if any(token in normalized for token in ("оператор", "оператором", "человеком", "менеджер", "сотрудник")):
return "operator_request"
if any(token in normalized for token in ("график", "распис", "время работы", "work schedule", "жұмыс")):
@@ -347,6 +373,8 @@ class AudioSocketMediaRuntime:
@staticmethod
def _ack_kind_for_intent(intent: str) -> str:
if intent == "closing":
return "closing"
if intent == "operator_request":
return "handoff"
if intent in {"schedule", "address", "price", "status", "problem"}:
@@ -406,8 +434,18 @@ class AudioSocketMediaRuntime:
return True
return normalized_intent != "unknown"
def _should_emit_blind_ack(self, actor: MediaActor, pcm_bytes: bytes) -> bool:
def _should_emit_blind_ack(self, actor: MediaActor, pcm_bytes: bytes, partial_transcript: str) -> bool:
"""Duration-only fallback for when no usable partial transcript exists yet.
Must defer to the transcript when one *is* available: otherwise a caller
who already said a recognized filler-answer ("да"/"нет"/"хорошо") still
gets a blind ack just because the audio happened to cross the length
threshold, even though `_should_emit_partial_ack` correctly said no.
"""
del actor
transcript_text = str(partial_transcript or "").strip()
if transcript_text and self._is_low_signal_partial_transcript(transcript_text):
return False
return len(pcm_bytes) >= self._immediate_ack_min_bytes
@staticmethod
@@ -1116,10 +1154,22 @@ class AudioSocketMediaRuntime:
metadata: dict[str, Any],
ack_source: str,
ack_kind: str | None = None,
intent: str | None = None,
) -> None:
if actor.closed or actor.early_ack_started:
return
ack_kind = ack_kind or self._ack_kind_for_intent(actor.partial_intent or "unknown")
effective_intent = str(
intent or actor.stable_partial_intent or actor.partial_intent or "unknown"
).strip() or "unknown"
if effective_intent == "closing":
# The caller is wrapping up; a "thinking" filler right before the
# closing reply reads as robotic, so skip it and go straight to the reply.
return
if actor.last_ack_completed_monotonic and (
time.monotonic() - actor.last_ack_completed_monotonic
) < self._ack_min_repeat_gap_seconds:
return
ack_kind = ack_kind or self._ack_kind_for_intent(effective_intent)
ack_text, style_hints, ack_variant = self._select_ack_payload(
actor,
language=language,
@@ -1620,8 +1670,9 @@ class AudioSocketMediaRuntime:
language=actor.registration.language,
metadata=base_metadata,
ack_source="streaming_partial" if actor.asr_streaming_enabled else "precomputed_partial_asr",
intent=partial_intent,
)
elif self._should_emit_blind_ack(actor, pcm_bytes):
elif self._should_emit_blind_ack(actor, pcm_bytes, partial_transcript):
await self._emit_early_ack(
actor,
language=actor.registration.language,
@@ -114,6 +114,10 @@ class RuntimeConfiguredTTSProvider(TTSProvider):
self._provider_cache[cache_key] = provider
return provider
def cache_fingerprint(self, language: str | None, *, style_hints: dict[str, object] | None = None) -> str:
provider = self._provider_for_language(language)
return f"{provider.name}:{provider.cache_fingerprint(language, style_hints=style_hints)}"
def synthesize(self, text: str, *, language: str | None = None, style_hints: dict[str, object] | None = None):
provider = self._provider_for_language(language)
return provider.synthesize(text, language=language, style_hints=style_hints)
+11 -6
View File
@@ -412,13 +412,18 @@ def ami_loop(stop_event=None) -> None:
raise RuntimeError("AMI connection closed")
if not frame:
continue
if frame.get("Event") != "UserEvent":
event_name = frame.get("Event")
if event_name == "UserEvent":
user_event = str(frame.get("UserEvent") or "").strip()
if not user_event.startswith(bridge._ami_prefix()):
continue
bridge._STATE.set_last_event()
bridge._record_ami_payload(frame)
elif event_name in {"DialEnd", "Hangup"}:
bridge._STATE.set_last_event()
bridge._record_ami_payload(frame, event_name=event_name)
else:
continue
user_event = str(frame.get("UserEvent") or "").strip()
if not user_event.startswith(bridge._ami_prefix()):
continue
bridge._STATE.set_last_event()
bridge._record_ami_payload(frame)
except Exception as exc:
bridge._STATE.set_connected(False)
bridge._STATE.set_error(str(exc))
+6
View File
@@ -262,6 +262,7 @@ def _start_background_threads() -> None:
extra_loops: list[tuple[str, Callable[..., Any]]] = []
if _ivr_fastagi_enabled():
extra_loops.append(("asterisk-ivr-fastagi-loop", _ivr_fastagi_loop))
extra_loops.append(("agent-acw-sweep-loop", _acw_sweep_loop))
bridge_runtime.start_background_threads(
ami_loop=_ami_loop,
failed_retry_loop=_failed_retry_loop,
@@ -305,6 +306,9 @@ _update_voice_ai_call_state = bridge_voice_ai.update_call_ai_state
_voice_ai_summary_for_call = bridge_voice_ai.voice_ai_summary_for_call
_create_escalation = bridge_voice_ai.create_escalation
_release_routing_agent = bridge_voice_ai.release_routing_agent
_retry_escalation_no_answer = bridge_voice_ai.retry_escalation_no_answer
_routing_release_by_agent_id = bridge_voice_ai.routing_release_by_agent_id
_set_routing_agent_status = bridge_voice_ai.set_routing_agent_status
_first_non_empty = bridge_ami.first_non_empty
_extract_call_id = bridge_ami.extract_call_id
@@ -373,6 +377,8 @@ _process_audio_bridge_ended = bridge_processing.process_audio_bridge_ended
_process_call_ended = bridge_processing.process_call_ended
_process_operator_connected = bridge_processing.process_operator_connected
_process_recording_ready = bridge_processing.process_recording_ready
_process_agent_dial_outcome = bridge_processing.process_agent_dial_outcome
_acw_sweep_loop = bridge_processing.acw_sweep_loop
_process_bridge_row = bridge_processing.process_bridge_row
_record_ami_payload = bridge_processing.record_ami_payload
_retry_failed_events_once = bridge_processing.retry_failed_events_once
@@ -4,6 +4,7 @@ from datetime import datetime, timezone
import hashlib
import json
import logging
import os
from pathlib import Path
import threading
from typing import Any
@@ -884,18 +885,34 @@ def process_call_ended(
open_escalation = session.execute(
select(EscalationRow).where(
EscalationRow.call_id == row.call_id,
EscalationRow.status.in_(["requested", "ringing"]),
EscalationRow.status.in_(["requested", "ringing", "connected"]),
)
).scalar_one_or_none()
was_talking = open_escalation is not None and open_escalation.status == "connected"
if open_escalation is not None:
open_escalation.status = "completed" if answered else "failed"
open_escalation.completed_at = now
if answered and not open_escalation.connected_at:
open_escalation.connected_at = now
try:
bridge._release_routing_agent(row.call_id)
except Exception:
pass
if was_talking and open_escalation and open_escalation.real_agent_id:
try:
bridge._set_routing_agent_status(open_escalation.real_agent_id, "AFTER_CALL_WORK")
except Exception:
logger.warning("bridge.agent_acw_status_failed call_id=%s agent_id=%s", row.call_id, open_escalation.real_agent_id)
else:
try:
bridge._release_routing_agent(row.call_id)
except Exception:
pass
if open_escalation is not None:
try:
bridge._append_interaction_timeline(
interaction_id=link.interaction_id,
action="escalation.completed" if answered else "escalation.failed_client_disconnected",
metadata={"call_id": row.call_id, "escalation_id": open_escalation.escalation_id, "agent_id": open_escalation.real_agent_id},
)
except Exception:
pass
try:
bridge._notify_voice_ai_telephony_event(
voice_session_id=link.voice_session_id,
@@ -969,6 +986,48 @@ def process_operator_connected(
link.telephony_status = "connected"
link.connected_at = now
link.updated_at = now
open_escalation = session.execute(
select(EscalationRow).where(
EscalationRow.call_id == row.call_id,
EscalationRow.status == "ringing",
)
).scalar_one_or_none()
if open_escalation is not None:
open_escalation.status = "connected"
open_escalation.connected_at = now
if open_escalation.real_agent_id:
try:
bridge._set_routing_agent_status(open_escalation.real_agent_id, "TALKING")
except Exception:
logger.warning("bridge.agent_talking_status_failed call_id=%s agent_id=%s", row.call_id, open_escalation.real_agent_id)
try:
bridge._append_interaction_timeline(
interaction_id=link.interaction_id,
action="escalation.agent_connected",
metadata={"call_id": row.call_id, "agent_id": open_escalation.real_agent_id, "escalation_id": open_escalation.escalation_id},
)
except Exception:
pass
try:
bridge._emit_voice_event(
event_type="AgentConnected",
call_id=row.call_id,
interaction_id=link.interaction_id,
payload={"escalation_id": open_escalation.escalation_id, "agent_id": open_escalation.real_agent_id},
)
except Exception:
pass
try:
bridge._emit_voice_event(
event_type="TransferCompleted",
call_id=row.call_id,
interaction_id=link.interaction_id,
payload={"escalation_id": open_escalation.escalation_id, "agent_id": open_escalation.real_agent_id},
)
except Exception:
pass
_upsert_voice_reporting_fact(
session,
link,
@@ -1173,6 +1232,55 @@ def process_recording_ready(
local_path.unlink(missing_ok=True)
_NO_ANSWER_DIAL_STATUSES = {"NOANSWER", "BUSY", "CANCEL", "CHANUNAVAIL", "CONGESTION"}
_NO_ANSWER_HANGUP_CAUSES = {"17", "18", "19", "21", "34", "38"}
def process_agent_dial_outcome(session, row: AsteriskEventLogRow, payload: dict[str, Any]) -> None:
bridge = _bridge_app()
call_id = row.call_id
if not call_id:
bridge._mark_log_forwarded(session, row, interaction_id=None)
return
dial_status = str(payload.get("DialStatus") or "").strip().upper()
hangup_cause = str(payload.get("Cause") or "").strip()
is_no_answer_outcome = (
(row.ami_event_name == "DialEnd" and dial_status in _NO_ANSWER_DIAL_STATUSES)
or (row.ami_event_name == "Hangup" and hangup_cause in _NO_ANSWER_HANGUP_CAUSES)
)
if not is_no_answer_outcome:
bridge._mark_log_forwarded(session, row, interaction_id=None)
return
escalation = session.execute(
select(EscalationRow).where(
EscalationRow.call_id == call_id,
EscalationRow.status == "ringing",
)
).scalar_one_or_none()
if escalation is None:
bridge._mark_log_forwarded(session, row, interaction_id=None)
return
dialed_channel = payload.get("DestChannel") if row.ami_event_name == "DialEnd" else payload.get("Channel")
dialed_extension = bridge._extract_extension_from_channel(dialed_channel)
if dialed_extension and escalation.assigned_agent_id and dialed_extension != escalation.assigned_agent_id:
bridge._mark_log_forwarded(session, row, interaction_id=None)
return
try:
bridge._retry_escalation_no_answer(
session,
call_id=call_id,
dial_outcome=dial_status or f"hangup_cause_{hangup_cause}",
)
except Exception:
logger.exception("bridge.agent_dial_outcome_retry_failed call_id=%s", call_id)
bridge._mark_log_forwarded(session, row, interaction_id=None)
def process_bridge_row(session, row: AsteriskEventLogRow) -> AsteriskEventLogRow:
bridge = _bridge_app()
payload = json.loads(row.payload_json or "{}")
@@ -1209,6 +1317,8 @@ def process_bridge_row(session, row: AsteriskEventLogRow) -> AsteriskEventLogRow
bridge._process_call_ended(session, row, payload)
elif row.ami_event_name == f"{bridge._ami_prefix()}RecordingReady":
bridge._process_recording_ready(session, row, payload)
elif row.ami_event_name in {"DialEnd", "Hangup"}:
bridge._process_agent_dial_outcome(session, row, payload)
else:
row.forward_status = "received"
row.updated_at = utc_now_iso()
@@ -1216,9 +1326,9 @@ def process_bridge_row(session, row: AsteriskEventLogRow) -> AsteriskEventLogRow
return row
def record_ami_payload(payload: dict[str, Any]) -> None:
def record_ami_payload(payload: dict[str, Any], *, event_name: str | None = None) -> None:
bridge = _bridge_app()
event_name = str(payload.get("UserEvent") or "").strip()
event_name = str(event_name or payload.get("UserEvent") or "").strip()
call_id = bridge._extract_call_id(payload)
linked_id = bridge._extract_linked_id(payload, call_id)
if not event_name or not call_id:
@@ -1262,6 +1372,42 @@ def retry_failed_events_once() -> None:
bridge._process_claimed_bridge_event(bridge_event_id)
def acw_duration_seconds() -> int:
raw = str(os.environ.get("ACW_DURATION_SECONDS", "30")).strip()
try:
return max(int(raw), 1)
except ValueError:
return 30
def acw_sweep_interval_seconds() -> float:
return max(min(float(acw_duration_seconds()) / 2, 15.0), 5.0)
def acw_sweep_once() -> None:
from services.routing_service import engine as routing_engine
session = get_session()
try:
swept = routing_engine.sweep_after_call_work(session, older_than_seconds=acw_duration_seconds())
for agent in swept:
logger.warning("bridge.agent_acw_swept agent_id=%s", agent.agent_id)
finally:
session.close()
def acw_sweep_loop(stop_event: threading.Event | None = None) -> None:
bridge = _bridge_app()
active_stop_event = stop_event or bridge._background_stop_event()
while not active_stop_event.is_set():
try:
acw_sweep_once()
except Exception:
logger.exception("bridge.acw_sweep_failed")
if active_stop_event.wait(acw_sweep_interval_seconds()):
break
def failed_retry_loop(stop_event: threading.Event | None = None) -> None:
bridge = _bridge_app()
active_stop_event = stop_event or bridge._background_stop_event()
+216 -18
View File
@@ -63,6 +63,57 @@ def append_interaction_timeline(
)
def _append_escalation_timeline(
escalation: EscalationRow,
link: AsteriskCallLinkRow,
*,
action: str,
extra: dict[str, Any] | None = None,
) -> None:
try:
append_interaction_timeline(
interaction_id=link.interaction_id,
action=action,
metadata={
"call_id": escalation.call_id,
"escalation_id": escalation.escalation_id,
"from_level": escalation.from_level,
"to_level": escalation.to_level,
"attempt_count": escalation.attempt_count,
**(extra or {}),
},
)
except Exception:
pass
def _emit_escalation_event(
*,
event_type: str,
escalation: EscalationRow,
link: AsteriskCallLinkRow,
extra: dict[str, Any] | None = None,
) -> None:
bridge = _bridge_app()
try:
bridge._emit_voice_event(
event_type=event_type,
call_id=escalation.call_id,
interaction_id=link.interaction_id,
payload={
"escalation_id": escalation.escalation_id,
"tenant_id": escalation.tenant_id,
"from_level": escalation.from_level,
"to_level": escalation.to_level,
"reason_code": escalation.reason_code,
"attempt_count": escalation.attempt_count,
**(extra or {}),
},
)
except Exception:
LOGGER.warning("bridge.escalation_event_failed event_type=%s call_id=%s", event_type, escalation.call_id)
def start_voice_ai_session(
*,
call_id: str,
@@ -188,7 +239,8 @@ def _reserve_routing_agent(
level: str,
tenant_id: str | None,
required_skills: list[str] | None = None,
) -> str | None:
exclude_agent_ids: list[str] | None = None,
) -> dict[str, Any] | None:
bridge = _bridge_app()
try:
response = bridge._post_json(
@@ -198,7 +250,7 @@ def _reserve_routing_agent(
"level": level,
"tenant_id": tenant_id,
"required_skills": required_skills or [],
"exclude_agent_ids": [],
"exclude_agent_ids": exclude_agent_ids or [],
},
timeout_seconds=bridge._callcontrol_side_effect_timeout_seconds(),
max_attempts=1,
@@ -212,8 +264,43 @@ def _reserve_routing_agent(
tenant_id,
)
return None
agent_id = str((response or {}).get("agent_id") or "").strip()
extension = str((response or {}).get("extension") or "").strip()
return extension or None
if not agent_id or not extension:
return None
return {
"agent_id": agent_id,
"extension": extension,
"endpoint": (response or {}).get("endpoint"),
"display_name": (response or {}).get("display_name"),
}
def set_routing_agent_status(agent_id: str, status: str) -> None:
bridge = _bridge_app()
try:
bridge._patch_json(
f"{bridge._routing_service_url()}/internal/routing/agents/{agent_id}/status",
{"status": status},
timeout_seconds=bridge._callcontrol_side_effect_timeout_seconds(),
max_attempts=1,
retry_backoff_seconds=0.0,
)
except Exception:
LOGGER.warning("bridge.routing_set_status_failed agent_id=%s status=%s", agent_id, status)
def _redirect_channel_to_agent(*, channel: str, extension: str) -> None:
bridge = _bridge_app()
bridge._ami_action(
"Redirect",
{
"Channel": channel,
"Context": bridge._transfer_context(),
"Exten": extension,
"Priority": 1,
},
)
def _resolve_handoff_extension(
@@ -233,14 +320,14 @@ def _resolve_handoff_extension(
level = target_level or bridge._routing_level_for_queue_code(queue_code)
if level and call_id:
resolved_tenant_id = tenant_id if tenant_id is not None else bridge._routing_tenant_for_queue_code(queue_code)
reserved_extension = _reserve_routing_agent(
reserved_agent = _reserve_routing_agent(
call_id=call_id,
level=level,
tenant_id=resolved_tenant_id,
required_skills=required_skills,
)
if reserved_extension:
return queue_code, reserved_extension, level, resolved_tenant_id
if reserved_agent:
return queue_code, reserved_agent["extension"], level, resolved_tenant_id
raise HTTPException(status_code=409, detail=f"No available {level} agent right now")
extension = bridge._transfer_target_map().get(queue_code)
@@ -666,6 +753,7 @@ def _escalation_to_out(row: EscalationRow) -> EscalationOut:
summary=row.summary,
status=row.status,
assigned_agent_id=row.assigned_agent_id,
attempt_count=row.attempt_count or 0,
requested_at=row.requested_at,
connected_at=row.connected_at,
completed_at=row.completed_at,
@@ -709,28 +797,32 @@ def create_escalation(call_id: str, body: EscalationRequestIn, actor: dict) -> E
session.flush()
channel = _resolve_handoff_channel(session, link)
reserved_extension = _reserve_routing_agent(
agent = _reserve_routing_agent(
call_id=call_id,
level=body.target_level,
tenant_id=link.tenant_id,
required_skills=body.required_skills,
)
if not reserved_extension:
if not agent:
escalation.status = "failed"
escalation.completed_at = now
session.commit()
_emit_escalation_event(
event_type="TransferFailed",
escalation=escalation,
link=link,
extra={"reason": "no_available_agent"},
)
raise HTTPException(status_code=409, detail=f"No available {body.target_level} agent right now")
escalation.real_agent_id = agent["agent_id"]
escalation.attempted_agent_ids_json = json.dumps([agent["agent_id"]], ensure_ascii=False)
session.flush()
_append_escalation_timeline(escalation, link, action="escalation.agent_reserved", extra={"agent_id": agent["agent_id"]})
_emit_escalation_event(event_type="AgentReserved", escalation=escalation, link=link, extra={"agent_id": agent["agent_id"]})
try:
bridge._ami_action(
"Redirect",
{
"Channel": channel,
"Context": bridge._transfer_context(),
"Exten": reserved_extension,
"Priority": 1,
},
)
_redirect_channel_to_agent(channel=channel, extension=agent["extension"])
except Exception as exc:
try:
bridge._release_routing_agent(call_id)
@@ -739,10 +831,13 @@ def create_escalation(call_id: str, body: EscalationRequestIn, actor: dict) -> E
escalation.status = "failed"
escalation.completed_at = utc_now_iso()
session.commit()
_append_escalation_timeline(escalation, link, action="escalation.transfer_failed", extra={"agent_id": agent["agent_id"], "error": str(exc)})
_emit_escalation_event(event_type="TransferFailed", escalation=escalation, link=link, extra={"agent_id": agent["agent_id"], "error": str(exc)})
raise HTTPException(status_code=502, detail=f"Failed to redirect call to agent: {exc}") from exc
set_routing_agent_status(agent["agent_id"], "RINGING")
escalation.status = "ringing"
escalation.assigned_agent_id = reserved_extension
escalation.assigned_agent_id = agent["extension"]
link.current_level = body.target_level
link.required_skills_json = json.dumps(body.required_skills, ensure_ascii=False)
link.priority = body.priority
@@ -751,11 +846,114 @@ def create_escalation(call_id: str, body: EscalationRequestIn, actor: dict) -> E
link.operator_extension = None
link.updated_at = now
session.commit()
_append_escalation_timeline(escalation, link, action="escalation.agent_ringing", extra={"agent_id": agent["agent_id"]})
_emit_escalation_event(event_type="AgentRinging", escalation=escalation, link=link, extra={"agent_id": agent["agent_id"]})
return _escalation_to_out(escalation)
finally:
session.close()
def retry_escalation_no_answer(session, *, call_id: str, dial_outcome: str) -> None:
"""AC-08 / ТЗ §14: агент не ответил — освободить его и попробовать следующего.
Вызывается из bridge_processing при нативном AMI DialEnd/Hangup с исходом
NOANSWER/BUSY/CANCEL/CHANUNAVAIL/CONGESTION на канале агента, зарезервированного
под активную (status='ringing') эскалацию этого call_id.
"""
bridge = _bridge_app()
escalation = session.execute(
select(EscalationRow).where(
EscalationRow.call_id == call_id,
EscalationRow.status == "ringing",
)
).scalar_one_or_none()
if escalation is None:
return
link = bridge._find_call_link(session, call_id)
if link is None:
return
now = utc_now_iso()
no_answer_agent_id = escalation.real_agent_id
if no_answer_agent_id:
try:
routing_release_by_agent_id(no_answer_agent_id)
except Exception:
LOGGER.warning("bridge.escalation_retry_release_failed call_id=%s agent_id=%s", call_id, no_answer_agent_id)
escalation.attempt_count = (escalation.attempt_count or 0) + 1
session.flush()
_append_escalation_timeline(
escalation,
link,
action="escalation.agent_no_answer",
extra={"agent_id": no_answer_agent_id, "dial_outcome": dial_outcome},
)
_emit_escalation_event(
event_type="AgentNoAnswer",
escalation=escalation,
link=link,
extra={"agent_id": no_answer_agent_id, "dial_outcome": dial_outcome},
)
attempted_ids = json.loads(escalation.attempted_agent_ids_json or "[]")
channel = _resolve_handoff_channel(session, link)
required_skills = json.loads(escalation.required_skills_json or "[]")
next_agent = _reserve_routing_agent(
call_id=call_id,
level=escalation.to_level,
tenant_id=escalation.tenant_id,
required_skills=required_skills,
exclude_agent_ids=attempted_ids,
)
if not next_agent:
escalation.status = "failed"
escalation.completed_at = now
session.commit()
_append_escalation_timeline(escalation, link, action="escalation.failed_no_agents")
_emit_escalation_event(event_type="TransferFailed", escalation=escalation, link=link, extra={"reason": "no_more_agents"})
LOGGER.warning("bridge.escalation_retry_exhausted call_id=%s attempts=%s", call_id, escalation.attempt_count)
return
escalation.real_agent_id = next_agent["agent_id"]
escalation.attempted_agent_ids_json = json.dumps(attempted_ids + [next_agent["agent_id"]], ensure_ascii=False)
session.flush()
_append_escalation_timeline(escalation, link, action="escalation.agent_reserved", extra={"agent_id": next_agent["agent_id"]})
_emit_escalation_event(event_type="AgentReserved", escalation=escalation, link=link, extra={"agent_id": next_agent["agent_id"]})
try:
_redirect_channel_to_agent(channel=channel, extension=next_agent["extension"])
except Exception as exc:
try:
routing_release_by_agent_id(next_agent["agent_id"])
except Exception:
pass
escalation.status = "failed"
escalation.completed_at = utc_now_iso()
session.commit()
_append_escalation_timeline(escalation, link, action="escalation.transfer_failed", extra={"agent_id": next_agent["agent_id"], "error": str(exc)})
_emit_escalation_event(event_type="TransferFailed", escalation=escalation, link=link, extra={"agent_id": next_agent["agent_id"], "error": str(exc)})
return
set_routing_agent_status(next_agent["agent_id"], "RINGING")
escalation.assigned_agent_id = next_agent["extension"]
session.commit()
_append_escalation_timeline(escalation, link, action="escalation.agent_ringing", extra={"agent_id": next_agent["agent_id"]})
_emit_escalation_event(event_type="AgentRinging", escalation=escalation, link=link, extra={"agent_id": next_agent["agent_id"]})
def routing_release_by_agent_id(agent_id: str, *, next_status: str = "AVAILABLE") -> None:
bridge = _bridge_app()
bridge._post_json(
f"{bridge._routing_service_url()}/internal/routing/release-agent",
{"agent_id": agent_id, "next_status": next_status},
timeout_seconds=bridge._callcontrol_side_effect_timeout_seconds(),
max_attempts=1,
retry_backoff_seconds=0.0,
)
def update_call_ai_state(call_id: str, body: VoiceAICallStateUpdateIn, actor: dict) -> VoiceLiveCallOut:
bridge = _bridge_app()
assert_trusted_voice_runtime_actor(actor)
+27 -4
View File
@@ -18,6 +18,7 @@ from services.shared.models import (
QueueOut,
RoutingAgentReserveIn,
RoutingAgentReserveOut,
RoutingAgentStatusIn,
)
from services.shared.security import require_roles
from services.shared.sql_init import init_sql_schema
@@ -343,6 +344,7 @@ def _escalation_to_out(row: EscalationRow) -> EscalationOut:
summary=row.summary,
status=row.status,
assigned_agent_id=row.assigned_agent_id,
attempt_count=row.attempt_count or 0,
requested_at=row.requested_at,
connected_at=row.connected_at,
completed_at=row.completed_at,
@@ -399,11 +401,32 @@ def release_agent_endpoint(
_: dict = Depends(require_roles(Role.ADMIN)),
) -> dict:
call_id = str(payload.get("call_id") or "").strip()
if not call_id:
raise HTTPException(status_code=400, detail="call_id is required")
agent_id = str(payload.get("agent_id") or "").strip()
next_status = str(payload.get("next_status") or "AVAILABLE").strip() or "AVAILABLE"
if not call_id and not agent_id:
raise HTTPException(status_code=400, detail="call_id or agent_id is required")
session = get_session()
try:
agent = routing_engine.release_agent_by_call_id(session, call_id=call_id)
return {"call_id": call_id, "released_agent_id": agent.agent_id if agent else None}
if agent_id:
agent = routing_engine.release_agent_by_id(session, agent_id=agent_id, next_status=next_status)
else:
agent = routing_engine.release_agent_by_call_id(session, call_id=call_id)
return {"call_id": call_id or None, "released_agent_id": agent.agent_id if agent else None}
finally:
session.close()
@app.patch("/internal/routing/agents/{agent_id}/status")
def set_agent_status_endpoint(
agent_id: str,
payload: RoutingAgentStatusIn,
_: dict = Depends(require_roles(Role.ADMIN)),
) -> dict:
session = get_session()
try:
agent = routing_engine.set_agent_status(session, agent_id=agent_id, status=payload.status)
if agent is None:
raise HTTPException(status_code=404, detail="Agent not found")
return {"agent_id": agent.agent_id, "status": agent.status}
finally:
session.close()
+36
View File
@@ -1,6 +1,7 @@
from __future__ import annotations
import json
from datetime import datetime
from sqlalchemy import select, text
@@ -117,6 +118,41 @@ def release_agent_by_call_id(session, *, call_id: str) -> AgentRow | None:
return agent
def set_agent_status(session, *, agent_id: str, status: str) -> AgentRow | None:
agent = session.execute(
select(AgentRow).where(AgentRow.agent_id == agent_id)
).scalar_one_or_none()
if agent is None:
return None
agent.status = status
agent.updated_at = utc_now_iso()
session.commit()
return agent
def sweep_after_call_work(session, *, older_than_seconds: int) -> list[AgentRow]:
cutoff = utc_now_iso()
rows = session.execute(
select(AgentRow).where(AgentRow.status == "AFTER_CALL_WORK")
).scalars().all()
swept: list[AgentRow] = []
for agent in rows:
try:
age_seconds = (
datetime.fromisoformat(cutoff) - datetime.fromisoformat(str(agent.updated_at))
).total_seconds()
except (TypeError, ValueError):
continue
if age_seconds >= older_than_seconds:
agent.status = "AVAILABLE"
agent.current_call_id = None
agent.updated_at = cutoff
swept.append(agent)
if swept:
session.commit()
return swept
def release_agent_by_id(session, *, agent_id: str, next_status: str = "AVAILABLE") -> AgentRow | None:
agent = session.execute(
select(AgentRow).where(AgentRow.agent_id == agent_id)
+40 -13
View File
@@ -1,6 +1,7 @@
from __future__ import annotations
import json
import os
from typing import Any
from sqlalchemy import select
@@ -13,22 +14,48 @@ from services.shared.sql_models import AIOperatorSettingsRow
AI_OPERATOR_SETTINGS_KEY = "global"
def _env_field(name: str, default: str) -> str:
value = os.getenv(name)
if value is None:
return default
normalized = value.strip()
return normalized or default
def ai_operator_default_config() -> AIOperatorConfig:
agent_name = _env_field("AI_OPERATOR_AGENT_NAME", "Айнур")
company_name = _env_field("AI_OPERATOR_COMPANY_NAME", "DigiOps")
return AIOperatorConfig(
agent_name="Айнур",
company_name="DigiOps",
base_system_prompt=(
"Ты Айнур, единый ИИ-оператор контакт-центра DigiOps для звонков, Telegram и других каналов. "
"Всегда сохраняй одну и ту же личность: тебя зовут Айнур. Если клиент спрашивает, кто ты или как "
"тебя зовут, отвечай, что ты Айнур. Отвечай естественно, кратко и по делу. Когда говоришь о себе, "
"используй женский род: могла, смогла, сделала, готова, проверила, нашла. Не завершай диалог "
"самостоятельно и говори «до свидания» только если клиент явно попрощался или попросил завершить "
"разговор. Телефонные номера читай по цифрам. Не используй Markdown, URL или таблицы."
agent_name=agent_name,
company_name=company_name,
base_system_prompt=_env_field(
"AI_OPERATOR_BASE_SYSTEM_PROMPT",
(
f"Ты {agent_name}, единый ИИ-оператор контакт-центра {company_name} для звонков, Telegram и других "
f"каналов. Всегда сохраняй одну и ту же личность: тебя зовут {agent_name}. Если клиент спрашивает, "
f"кто ты или как тебя зовут, отвечай, что ты {agent_name}. Отвечай естественно, кратко и по делу. "
"Когда говоришь о себе, используй женский род: могла, смогла, сделала, готова, проверила, нашла. "
"Не завершай диалог самостоятельно и говори «до свидания» только если клиент явно попрощался или "
"попросил завершить разговор. Телефонные номера читай по цифрам. Не используй Markdown, URL или "
"таблицы."
),
),
identity_reply_ru=_env_field(
"AI_OPERATOR_IDENTITY_REPLY_RU",
f"Я {agent_name}, оператор контакт-центра {company_name}. Чем могу помочь?",
),
identity_reply_kz=_env_field(
"AI_OPERATOR_IDENTITY_REPLY_KZ",
f"Мен {agent_name}мын, {company_name} байланыс орталығының операторымын. Қалай көмектесе аламын?",
),
voice_greeting_ru=_env_field(
"AI_OPERATOR_VOICE_GREETING_RU",
f"Здравствуйте. Я {agent_name}. Подскажите, пожалуйста, чем помочь. (тест деплоя)",
),
voice_greeting_kz=_env_field(
"AI_OPERATOR_VOICE_GREETING_KZ",
f"Сәлеметсіз бе. Мен {agent_name}мын. Қалай көмектесе аламын?",
),
identity_reply_ru="Я Айнур, оператор контакт-центра DigiOps. Чем могу помочь?",
identity_reply_kz="Мен Айнурмын, DigiOps байланыс орталығының операторымын. Қалай көмектесе аламын?",
voice_greeting_ru="Здравствуйте. Я Айнур. Подскажите, пожалуйста, чем помочь. (тест деплоя)",
voice_greeting_kz="Сәлеметсіз бе. Мен Айнурмын. Қалай көмектесе аламын?",
)
+5
View File
@@ -1311,11 +1311,16 @@ class EscalationOut(BaseModel):
summary: str | None = None
status: str
assigned_agent_id: str | None = None
attempt_count: int = 0
requested_at: str
connected_at: str | None = None
completed_at: str | None = None
class RoutingAgentStatusIn(BaseModel):
status: str = Field(min_length=1)
class RoutingAgentReserveIn(BaseModel):
call_id: str = Field(min_length=1)
level: AgentLevel
+3
View File
@@ -846,6 +846,9 @@ class EscalationRow(Base):
summary: Mapped[str | None] = mapped_column(Text, nullable=True)
status: Mapped[str] = mapped_column(String(32), index=True, default="requested")
assigned_agent_id: Mapped[str | None] = mapped_column(String(64), nullable=True, index=True)
real_agent_id: Mapped[str | None] = mapped_column(String(64), nullable=True, index=True)
attempted_agent_ids_json: Mapped[str] = mapped_column(Text, default="[]")
attempt_count: Mapped[int] = mapped_column(Integer, default=0)
requested_at: Mapped[str] = mapped_column(String(64), index=True)
connected_at: Mapped[str | None] = mapped_column(String(64), nullable=True)
completed_at: Mapped[str | None] = mapped_column(String(64), nullable=True)
+55
View File
@@ -1681,6 +1681,61 @@ def test_voice_llm_prompt_includes_context_summary_and_uses_12_segments():
assert payload["history"][0]["sequence_no"] == 4
def test_voice_llm_prompt_instructs_model_not_to_self_name_customer():
messages = voice_module._voice_llm_prompt_messages(
language="ru",
customer=None,
interaction=SimpleNamespace(
interaction_id="int_voice_prompt_name",
status="open",
queue_id="que_voice",
subject="schedule",
customer_id="cus_voice_prompt_name",
),
transcript_text="Мне нужен график работы",
transcript_window=[],
conversation_summary_text="",
kb_results=[],
name_value="Ернур",
name_status="name_obtained",
)
system_prompt = messages[0]["content"]
assert "addressing the customer by name" in system_prompt
def test_voice_reply_with_name_does_not_duplicate_inflected_name_form():
# The model may address the customer using a grammatically declined form of
# their name ("Данияре" instead of "Данияр"); an exact-token dedup check
# would miss this and prepend the name a second time.
reply = voice_module._voice_reply_with_name(
"ru", "Здравствуйте, Данияре! Чем могу помочь?", "Данияр"
)
assert reply == "Здравствуйте, Данияре! Чем могу помочь?"
# A reply with no mention of the customer's name still gets it prefixed once.
reply = voice_module._voice_reply_with_name("ru", "Чем могу помочь?", "Данияр")
assert reply == "Данияр, Чем могу помочь?"
def test_voice_reply_with_name_greet_mode_uses_one_of_two_fixed_forms():
# Regular turns (name already known): just the name, never a greeting word.
reply = voice_module._voice_reply_with_name("ru", "Чем могу помочь?", "Данияр", greet=False)
assert reply == "Данияр, Чем могу помочь?"
# The turn the name is first learned: exactly "Здравствуйте, {name}, ...".
reply = voice_module._voice_reply_with_name("ru", "Чем могу помочь?", "Данияр", greet=True)
assert reply == "Здравствуйте, Данияр, Чем могу помочь?"
reply = voice_module._voice_reply_with_name("kz", "Немен көмектесе аламын?", "Ерлан", greet=True)
assert reply == "Сәлеметсіз бе, Ерлан, Немен көмектесе аламын?"
# Still deduplicates even in greet mode if the model already named the customer.
reply = voice_module._voice_reply_with_name(
"ru", "Здравствуйте, Данияре! Чем могу помочь?", "Данияр", greet=True
)
assert reply == "Здравствуйте, Данияре! Чем могу помочь?"
def test_voice_postprocess_reply_uses_summary_context_when_raw_window_lost_topic():
reply_text = voice_module._voice_postprocess_reply_text(
language="ru",
+282
View File
@@ -1249,6 +1249,288 @@ def test_media_runtime_voice_v2_emits_blind_ack_on_first_turn_without_partial_si
assert speak_events[1][0] == "Подскажите подробнее, пожалуйста."
def test_media_runtime_voice_v2_skips_filler_ack_when_caller_says_goodbye():
planned: list[tuple[str, str, str, dict | None]] = []
class _GoodbyeASRProvider(ASRProvider):
name = "goodbye-asr"
def transcribe(self, audio_bytes: bytes, *, language_hint: str | None = None) -> ASRTranscription:
assert audio_bytes
return ASRTranscription(text="Спасибо, до свидания", language=language_hint or "ru", confidence=0.9)
runtime = AudioSocketMediaRuntime(
enabled=True,
host="127.0.0.1",
port=0,
frame_ms=20,
idle_timeout_seconds=2.0,
registration_wait_timeout_seconds=0.5,
min_speech_ms=40,
trailing_silence_ms=40,
max_turn_ms=400,
asr_provider=_GoodbyeASRProvider(),
tts_provider=_StubTTSProvider(),
load_registration_by_media_uuid=lambda value: None,
mark_media_connected=lambda session_id, value: None,
mark_media_ended=lambda session_id, reason: None,
touch_media_frame=lambda session_id: None,
set_state=lambda session_id, state, handoff_reason, metadata: None,
get_pending_greeting=lambda session_id: None,
mark_reply_delivered=lambda session_id, text, is_greeting: None,
plan_reply=lambda session_id, text, metadata, kind: planned.append((session_id, text, kind, metadata)),
process_turn=lambda session_id, transcript_text, language, barge_in, metadata: (
time.sleep(0.25)
or VoiceAITurnDecisionOut(
language=language or "ru",
intent="closing",
reply_text="Хорошо, всего доброго!",
confidence=0.9,
needs_handoff=False,
handoff_reason=None,
case_action="close",
kb_refs=[],
summary_text="call wrapped up",
model="stub-voice",
latency_ms=1,
status="active",
)
),
request_handoff=lambda session_id, customer_request_text, decision: None,
handle_media_error=lambda session_id, message, metadata: None,
)
async def _fake_speak_text(
current_actor,
text: str,
*,
is_greeting: bool,
style_hints: dict[str, object] | None = None,
) -> None:
del current_actor, is_greeting, style_hints
await asyncio.sleep(0)
runtime._speak_text = _fake_speak_text # type: ignore[method-assign]
pcm_frame = (1000).to_bytes(2, "little", signed=True) * 160
async def _scenario() -> None:
actor = MediaActor(
registration=MediaRegistration(
voice_session_id="avs_media_runtime_v2_goodbye",
call_id="call_media_runtime_v2_goodbye",
interaction_id="int_media_runtime_v2_goodbye",
ai_session_id="ais_media_runtime_v2_goodbye",
language="ru",
media_uuid=str(uuid.uuid4()),
queue_code="voice_lab_ai",
queue_id="que_voice_lab_ai",
agent_profile="voice_support",
voice_v2_enabled=True,
voice_v2_ack_mode="immediate_short",
voice_v2_streaming_tts=True,
voice_v2_partial_asr=False,
),
reader=asyncio.StreamReader(),
writer=None, # type: ignore[arg-type]
vad=EnergyVAD(frame_ms=20, min_speech_ms=40, trailing_silence_ms=40, max_turn_ms=400),
frame_ms=20,
frame_bytes=320,
)
actor.finalized_caller_turn_count = 1
await runtime._process_utterance(actor, pcm_frame, False)
asyncio.run(_scenario())
assert [item[2] for item in planned] == ["reply"]
assert planned[0][1] == "Хорошо, всего доброго!"
def test_media_runtime_voice_v2_throttles_repeated_filler_ack_within_gap():
planned: list[tuple[str, str, str, dict | None]] = []
class _SlowASRProvider(ASRProvider):
name = "slow-asr"
def transcribe(self, audio_bytes: bytes, *, language_hint: str | None = None) -> ASRTranscription:
assert audio_bytes
return ASRTranscription(text="Хочу узнать график работы", language=language_hint or "ru", confidence=0.9)
runtime = AudioSocketMediaRuntime(
enabled=True,
host="127.0.0.1",
port=0,
frame_ms=20,
idle_timeout_seconds=2.0,
registration_wait_timeout_seconds=0.5,
min_speech_ms=40,
trailing_silence_ms=40,
max_turn_ms=400,
asr_provider=_SlowASRProvider(),
tts_provider=_StubTTSProvider(),
load_registration_by_media_uuid=lambda value: None,
mark_media_connected=lambda session_id, value: None,
mark_media_ended=lambda session_id, reason: None,
touch_media_frame=lambda session_id: None,
set_state=lambda session_id, state, handoff_reason, metadata: None,
get_pending_greeting=lambda session_id: None,
mark_reply_delivered=lambda session_id, text, is_greeting: None,
plan_reply=lambda session_id, text, metadata, kind: planned.append((session_id, text, kind, metadata)),
process_turn=lambda session_id, transcript_text, language, barge_in, metadata: (
time.sleep(0.25)
or VoiceAITurnDecisionOut(
language=language or "ru",
intent="clarification",
reply_text="Подскажите, пожалуйста, какой город вас интересует?",
confidence=0.9,
needs_handoff=False,
handoff_reason=None,
case_action="keep_open",
kb_refs=[],
summary_text="reply ready",
model="stub-voice",
latency_ms=1,
status="active",
)
),
request_handoff=lambda session_id, customer_request_text, decision: None,
handle_media_error=lambda session_id, message, metadata: None,
)
async def _fake_speak_text(
current_actor,
text: str,
*,
is_greeting: bool,
style_hints: dict[str, object] | None = None,
) -> None:
del current_actor, is_greeting, style_hints
await asyncio.sleep(0)
runtime._speak_text = _fake_speak_text # type: ignore[method-assign]
pcm_frame = (1000).to_bytes(2, "little", signed=True) * 160
async def _scenario() -> None:
actor = MediaActor(
registration=MediaRegistration(
voice_session_id="avs_media_runtime_v2_throttle",
call_id="call_media_runtime_v2_throttle",
interaction_id="int_media_runtime_v2_throttle",
ai_session_id="ais_media_runtime_v2_throttle",
language="ru",
media_uuid=str(uuid.uuid4()),
queue_code="voice_lab_ai",
queue_id="que_voice_lab_ai",
agent_profile="voice_support",
voice_v2_enabled=True,
voice_v2_ack_mode="immediate_short",
voice_v2_streaming_tts=True,
voice_v2_partial_asr=False,
),
reader=asyncio.StreamReader(),
writer=None, # type: ignore[arg-type]
vad=EnergyVAD(frame_ms=20, min_speech_ms=40, trailing_silence_ms=40, max_turn_ms=400),
frame_ms=20,
frame_bytes=320,
)
actor.finalized_caller_turn_count = 1
await runtime._process_utterance(actor, pcm_frame, False)
runtime._reset_live_turn_state(actor)
await runtime._process_utterance(actor, pcm_frame, False)
asyncio.run(_scenario())
assert [item[2] for item in planned] == ["ack", "reply", "reply"]
def test_media_runtime_voice_v2_blind_ack_defers_to_known_low_signal_partial_transcript():
speak_events: list[str] = []
runtime = AudioSocketMediaRuntime(
enabled=True,
host="127.0.0.1",
port=0,
frame_ms=20,
idle_timeout_seconds=2.0,
registration_wait_timeout_seconds=0.5,
min_speech_ms=40,
trailing_silence_ms=40,
max_turn_ms=2000,
asr_provider=_StubASRProvider(),
tts_provider=_StubTTSProvider(),
load_registration_by_media_uuid=lambda value: None,
mark_media_connected=lambda session_id, value: None,
mark_media_ended=lambda session_id, reason: None,
touch_media_frame=lambda session_id: None,
set_state=lambda session_id, state, handoff_reason, metadata: None,
get_pending_greeting=lambda session_id: None,
mark_reply_delivered=lambda session_id, text, is_greeting: None,
plan_reply=lambda session_id, text, metadata, kind: None,
process_turn=lambda session_id, transcript_text, language, barge_in, metadata: VoiceAITurnDecisionOut(
language=language or "ru",
intent="clarification",
reply_text="Подскажите подробнее, пожалуйста.",
confidence=0.9,
needs_handoff=False,
handoff_reason=None,
case_action="keep_open",
kb_refs=[],
summary_text="reply ready",
model="stub-voice",
latency_ms=1,
status="active",
),
request_handoff=lambda session_id, customer_request_text, decision: None,
handle_media_error=lambda session_id, message, metadata: None,
)
async def _fake_speak_text(
current_actor,
text: str,
*,
is_greeting: bool,
style_hints: dict[str, object] | None = None,
) -> None:
del current_actor, is_greeting, style_hints
speak_events.append(text)
await asyncio.sleep(0)
runtime._speak_text = _fake_speak_text # type: ignore[method-assign]
pcm_frame = (1000).to_bytes(2, "little", signed=True) * 160
async def _scenario() -> None:
actor = MediaActor(
registration=MediaRegistration(
voice_session_id="avs_media_runtime_v2_low_signal_blind",
call_id="call_media_runtime_v2_low_signal_blind",
interaction_id="int_media_runtime_v2_low_signal_blind",
ai_session_id="ais_media_runtime_v2_low_signal_blind",
language="ru",
media_uuid=str(uuid.uuid4()),
queue_code="voice_lab_ai",
queue_id="que_voice_lab_ai",
agent_profile="voice_support",
voice_v2_enabled=True,
voice_v2_ack_mode="immediate_short",
voice_v2_streaming_tts=True,
voice_v2_partial_asr=True,
),
reader=asyncio.StreamReader(),
writer=None, # type: ignore[arg-type]
vad=EnergyVAD(frame_ms=20, min_speech_ms=40, trailing_silence_ms=40, max_turn_ms=2000),
frame_ms=20,
frame_bytes=320,
)
actor.finalized_caller_turn_count = 1
# A caller who already said a recognized filler-answer ("да") should not
# get a blind ack just because the audio clip crossed the length threshold.
actor.stable_partial_transcript = "да"
await runtime._process_utterance(actor, pcm_frame * 40, False)
asyncio.run(_scenario())
assert speak_events == ["Подскажите подробнее, пожалуйста."]
def test_media_runtime_low_signal_filter_catches_short_asr_noise():
assert AudioSocketMediaRuntime._is_low_signal_partial_transcript("Давай")
assert AudioSocketMediaRuntime._is_low_signal_partial_transcript("твой")
+28
View File
@@ -855,3 +855,31 @@ def test_push_voice_ai_telephony_event_call_ended_returns_detached_safe_payload(
}
assert closed == [("avs_runtime_call_ended", "call_ended")]
assert orchestrator_calls == [("POST", "/ai/voice/sessions/avs_runtime_call_ended/close")]
def test_media_registration_emotive_ack_rotation_is_not_gated_on_v2_queue_eligibility(monkeypatch):
# AI_VOICE_V2_QUEUE_CODES defaults to "voice_lab_ai" only, so a queue outside
# that allowlist runs the plain v1 ack path. Emotive-ack rotation and prebaked-ack
# caching must still apply there — otherwise every filler collapses to the single
# fixed "Секунду." fallback string instead of rotating, and gets synthesized live
# via ElevenLabs each time instead of served from the pre-cached ack bank, adding
# real TTS round-trip time to the one phrase whose job is to hide that latency.
monkeypatch.delenv("AI_VOICE_V2_QUEUE_CODES", raising=False)
monkeypatch.delenv("AI_VOICE_V2_EMOTIVE_ACK_ENABLED", raising=False)
monkeypatch.delenv("AI_VOICE_V2_PREBAKED_ACK_ENABLED", raising=False)
row = VoiceAISessionRow(
session_id="avs_runtime_emotive_ack_v1",
call_id="call_runtime_emotive_ack_v1",
interaction_id="int_runtime_emotive_ack_v1",
ai_session_id="ais_runtime_emotive_ack_v1",
language="ru",
media_uuid="media-emotive-ack-v1",
queue_id="que_not_v2_eligible",
agent_profile="voice_support",
)
registration = runtime_module._media_registration_from_row(row, queue_code="queue_outside_v2_allowlist")
assert registration.voice_v2_enabled is False
assert registration.voice_v2_emotive_ack is True
assert registration.voice_v2_prebaked_ack is True
+2 -1
View File
@@ -2195,6 +2195,7 @@ def test_disabled_bridge_startup_does_not_require_singleton_guard(monkeypatch):
monkeypatch.delenv("ASTERISK_BRIDGE_ENABLED", raising=False)
monkeypatch.setattr(bridge_module, "_ami_loop", _wait_until_stopped)
monkeypatch.setattr(bridge_module, "_failed_retry_loop", _wait_until_stopped)
monkeypatch.setattr(bridge_module, "_acw_sweep_loop", _wait_until_stopped)
monkeypatch.setattr(
bridge_module,
"_try_acquire_bridge_singleton_guard",
@@ -2202,7 +2203,7 @@ def test_disabled_bridge_startup_does_not_require_singleton_guard(monkeypatch):
)
bridge_module._startup()
assert len(bridge_module._background_threads_alive()) == 2
assert len(bridge_module._background_threads_alive()) == 3
def test_shutdown_releases_singleton_guard(monkeypatch):
+190
View File
@@ -0,0 +1,190 @@
import json
import pytest
import services.asterisk_bridge_service.voice_ai as voice_ai
from services.routing_service import engine as routing_engine
from services.shared.core import new_id, utc_now_iso
from services.shared.db import get_session
from services.shared.sql_init import init_sql_schema
from services.shared.sql_models import AgentRow, AsteriskCallLinkRow, EscalationRow
def _make_agent(session, *, extension: str, status: str = "AVAILABLE"):
now = utc_now_iso()
row = AgentRow(
agent_id=new_id("agt"),
tenant_ids_json="[]",
extension=extension,
endpoint=None,
display_name=f"Agent {extension}",
level="L2",
skills_json="[]",
status=status,
max_concurrent_calls=1,
enabled=True,
created_at=now,
updated_at=now,
)
session.add(row)
session.commit()
session.refresh(row)
return row
def _make_call_link(session, *, call_id: str):
now = utc_now_iso()
link = AsteriskCallLinkRow(
call_id=call_id,
linked_id=call_id,
queue_code="voice_lab_ai",
queue_id="que_test",
interaction_id="int_test",
status="active",
telephony_status="ringing",
channel_name=f"PJSIP/tele2-kazgaz-{call_id}",
current_level="L2",
started_at=now,
updated_at=now,
)
session.add(link)
session.commit()
session.refresh(link)
return link
def _make_ringing_escalation(session, *, call_id: str, agent: AgentRow):
now = utc_now_iso()
escalation = EscalationRow(
escalation_id=new_id("esc"),
call_id=call_id,
tenant_id=None,
from_level="L1",
to_level="L2",
reason_code="AI_UNABLE_TO_RESOLVE",
required_skills_json="[]",
priority=3,
status="ringing",
assigned_agent_id=agent.extension,
real_agent_id=agent.agent_id,
attempted_agent_ids_json=json.dumps([agent.agent_id]),
requested_at=now,
)
session.add(escalation)
session.commit()
session.refresh(escalation)
return escalation
def _patch_routing_over_http(monkeypatch, session):
"""Make voice_ai's HTTP-facing routing helpers operate on the same test session directly."""
def fake_reserve(*, call_id, level, tenant_id, required_skills=None, exclude_agent_ids=None):
agent = routing_engine.reserve_agent(
session,
call_id=call_id,
level=level,
tenant_id=tenant_id,
required_skills=required_skills,
exclude_agent_ids=exclude_agent_ids,
)
if agent is None:
return None
return {
"agent_id": agent.agent_id,
"extension": agent.extension,
"endpoint": agent.endpoint,
"display_name": agent.display_name,
}
def fake_release_by_agent_id(agent_id, *, next_status="AVAILABLE"):
routing_engine.release_agent_by_id(session, agent_id=agent_id, next_status=next_status)
def fake_set_status(agent_id, status):
routing_engine.set_agent_status(session, agent_id=agent_id, status=status)
monkeypatch.setattr(voice_ai, "_reserve_routing_agent", fake_reserve)
monkeypatch.setattr(voice_ai, "routing_release_by_agent_id", fake_release_by_agent_id)
monkeypatch.setattr(voice_ai, "set_routing_agent_status", fake_set_status)
monkeypatch.setattr(voice_ai, "_append_escalation_timeline", lambda *a, **k: None)
monkeypatch.setattr(voice_ai, "_emit_escalation_event", lambda *a, **k: None)
monkeypatch.setattr(voice_ai, "_resolve_handoff_channel", lambda session, link: link.channel_name)
def test_retry_escalation_no_answer_moves_to_next_available_agent(monkeypatch):
init_sql_schema()
redirected_to: list[str] = []
session = get_session()
try:
_patch_routing_over_http(monkeypatch, session)
monkeypatch.setattr(
voice_ai,
"_redirect_channel_to_agent",
lambda *, channel, extension: redirected_to.append(extension),
)
agent_a = _make_agent(session, extension="2001")
agent_b = _make_agent(session, extension="2002")
routing_engine.reserve_agent(session, call_id="call-retry-1", level="L2", tenant_id=None, required_skills=[])
link = _make_call_link(session, call_id="call-retry-1")
escalation = _make_ringing_escalation(session, call_id="call-retry-1", agent=agent_a)
voice_ai.retry_escalation_no_answer(session, call_id="call-retry-1", dial_outcome="NOANSWER")
session.refresh(escalation)
session.refresh(agent_a)
session.refresh(agent_b)
assert escalation.attempt_count == 1
assert escalation.status == "ringing"
assert escalation.real_agent_id == agent_b.agent_id
assert escalation.assigned_agent_id == agent_b.extension
assert json.loads(escalation.attempted_agent_ids_json) == [agent_a.agent_id, agent_b.agent_id]
assert agent_a.status == "AVAILABLE"
assert agent_a.current_call_id is None
assert agent_b.status == "RINGING"
assert agent_b.current_call_id == "call-retry-1"
assert redirected_to == [agent_b.extension]
finally:
session.close()
def test_retry_escalation_no_answer_exhausts_pool_marks_failed(monkeypatch):
init_sql_schema()
session = get_session()
try:
_patch_routing_over_http(monkeypatch, session)
monkeypatch.setattr(voice_ai, "_redirect_channel_to_agent", lambda **kwargs: None)
agent_a = _make_agent(session, extension="3001")
routing_engine.reserve_agent(session, call_id="call-retry-2", level="L2", tenant_id=None, required_skills=[])
link = _make_call_link(session, call_id="call-retry-2")
escalation = _make_ringing_escalation(session, call_id="call-retry-2", agent=agent_a)
voice_ai.retry_escalation_no_answer(session, call_id="call-retry-2", dial_outcome="NOANSWER")
session.refresh(escalation)
session.refresh(agent_a)
assert escalation.attempt_count == 1
assert escalation.status == "failed"
assert agent_a.status == "AVAILABLE"
finally:
session.close()
def test_retry_escalation_no_answer_ignores_calls_without_ringing_escalation(monkeypatch):
init_sql_schema()
called = []
session = get_session()
try:
_patch_routing_over_http(monkeypatch, session)
monkeypatch.setattr(voice_ai, "_redirect_channel_to_agent", lambda **kwargs: called.append(kwargs))
_make_call_link(session, call_id="call-no-escalation")
voice_ai.retry_escalation_no_answer(session, call_id="call-no-escalation", dial_outcome="NOANSWER")
assert called == []
finally:
session.close()
+48
View File
@@ -1,3 +1,4 @@
from datetime import datetime, timedelta, timezone
import json
from sqlalchemy import select
@@ -134,3 +135,50 @@ def test_reserve_agent_excludes_disabled_and_excluded_ids():
assert reserved.agent_id != excluded.agent_id
finally:
session.close()
def test_set_agent_status_transitions_ringing_to_talking():
init_sql_schema()
session = get_session()
try:
agent = _make_agent(session, level="L2")
routing_engine.reserve_agent(session, call_id="call-status", level="L2", tenant_id=None, required_skills=[])
ringing = routing_engine.set_agent_status(session, agent_id=agent.agent_id, status="RINGING")
assert ringing is not None
assert ringing.status == "RINGING"
talking = routing_engine.set_agent_status(session, agent_id=agent.agent_id, status="TALKING")
assert talking.status == "TALKING"
missing = routing_engine.set_agent_status(session, agent_id="unknown-agent", status="AVAILABLE")
assert missing is None
finally:
session.close()
def test_sweep_after_call_work_releases_only_expired_agents():
init_sql_schema()
session = get_session()
try:
stale = _make_agent(session, level="L2", status="AFTER_CALL_WORK")
fresh = _make_agent(session, level="L2", status="AFTER_CALL_WORK")
stale_ts = (datetime.now(timezone.utc) - timedelta(seconds=120)).replace(microsecond=0).isoformat()
stale.updated_at = stale_ts
stale.current_call_id = "call-stale"
session.commit()
swept = routing_engine.sweep_after_call_work(session, older_than_seconds=30)
swept_ids = {a.agent_id for a in swept}
assert stale.agent_id in swept_ids
assert fresh.agent_id not in swept_ids
session.refresh(stale)
session.refresh(fresh)
assert stale.status == "AVAILABLE"
assert stale.current_call_id is None
assert fresh.status == "AFTER_CALL_WORK"
finally:
session.close()