Consolida o trabalho desta branch de abril/2026 em um bloco pronto pra
testar em staging antes do merge pra main.
## Correções de memória semântica
- ExtractionService: Princípio Zero + Regra de Ouro (ação consumada vs intenção).
- Cenário Daniela_Reservas: Passo 0 de classificação (consulta/intenção/fora).
## Roleta da Sorte (end-to-end)
- Schema Supabase + 7 RPCs atômicas (server-side, idempotentes).
- Services: Offer, Redeem, WeeklyReport.
- Jobs: OfferRouletteJob (hook em ConfirmationService após Pix pago),
NotifyRevealed + Scheduler de fallback.
- Tool manual GenerateRoletaLinkTool + endpoint público /roleta/notify.
- Dashboard /captain/roleta com Resgate + Relatório + anomaly detection.
## Cenário Reclamacoes_Ouvidoria
- Triagem P1-P4, framework LAST, Three-level listening, Self-check.
- Sem compensação material, detecção de cliente frustrado eleva prioridade.
## Analytics
- Funil de conversão /captain/funnel: 5 etapas via regex, zero LLM.
- Detector de churn via ChurnOutreach* (cron dias úteis 10h-17h BRT).
## Trabalho pré-existente incluído
- Captain Executive Reports (ceo_digest, mattermost_delivery).
- get_reserva_preco_tool, Lifecycle ajustes, Reservations UI polimentos.
## Outros
- .gitignore: patterns pra credenciais.
- Migrations de scenarios idempotentes.
- i18n completa pt_BR+en pra roleta/funnel.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds concierge.* and reservation.* Liquid variables to agent_instructions
so Sofia's orchestrator_prompt receives unit persona/knowledge/variables
and reservation data resolved from conversation.custom_attributes.current_unit_id.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The [Galeria de Fotos] rules previously added to assistant_response_generator
only apply to the legacy V1 chat service. In V2 (captain_integration_v2),
scenario agents use scenario.liquid as their system prompt template, not
assistant_response_generator.
This adds conditional rules to scenario.liquid (matching the existing pattern
for faq_lookup and check_pix_payment) that activate for any scenario that
has the send_suite_images tool:
- Infer suite_category vs suite_number from context, no confirmation needed
- Never announce photo sending before the tool confirms images were found
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Adiciona SYSTEM_PROMPT_LEAK_PATTERNS no ResponseBuilderJob para detectar quando o LLM retornou o system prompt em vez de uma resposta ao cliente
- Filtra mensagens contaminadas do historico de conversas antes de enviar ao LLM (evita contaminacao em espiral)
- Adiciona guardrail no validate_message_content! que redireciona para handoff humano em caso de vazamento detectado
- Cria Captain::Errors::SystemPromptLeakError para tipagem do erro
- Atualiza assistant.liquid com tags INSTRUCOES_INTERNAS e REGRA CRITICA para instruir o LLM a nao reproduzir o system prompt como resposta
## Description
This PR sets up an `Enterprise::Railtie` to correctly register rake
tasks in the `enterprise` namespace.
Previously, rake tasks under `enterprise/lib/tasks` were being eagerly
loaded at Rails boot, causing `undefined method 'namespace'` errors.
With this change, rake tasks are now registered only in the rake
context, avoiding boot-time issues and ensuring they are discoverable
with `bin/rake -T`.
**Tasks added:**
* `search:all` → Reindex messages for all accounts
* `search:account[ID]` → Reindex messages for a specific account
Fixes: #12414
Co-authored-by: Sojan Jose <sojan@pepalo.com>
We now support searching within the actual message content, email
subject lines, and audio transcriptions. This enables a faster, more
accurate search experience going forward. Unlike the standard message
search, which is limited to the last 3 months, this search has no time
restrictions.
The search engine also accounts for small variations in queries. Minor
spelling mistakes, such as searching for slck instead of Slack, will
still return the correct results. It also ignores differences in accents
and diacritics, so searching for Deja vu will match content containing
Déjà vu.
We can also refine searches in the future by criteria such as:
- Searching within a specific inbox
- Filtering by sender or recipient
- Limiting to messages sent by an agent
Fixes https://github.com/chatwoot/chatwoot/issues/11656
Fixes https://github.com/chatwoot/chatwoot/issues/10669
Fixes https://github.com/chatwoot/chatwoot/issues/5910
---
Rake tasks to reindex all the messages.
```sh
bundle exec rake search:all
```
Rake task to reindex messages from one account only
```sh
bundle exec rake search:account ACCOUNT_ID=1
```
This PR migrates the legacy OpenAI integration (where users provide
their own API keys) from using hardcoded `https://api.openai.com`
endpoints to use the configurable `CAPTAIN_OPEN_AI_ENDPOINT` from the
captain configuration. This ensures consistency across all OpenAI
integrations in the platform.
## Changes
- Updated `lib/integrations/openai_base_service.rb` to use captain
endpoint config
- Updated `enterprise/app/models/enterprise/concerns/article.rb` to use
captain endpoint config
- Removed unused `enterprise/lib/chat_gpt.rb` class
- Added tests for endpoint configuration behavior
Migration Guide: https://chwt.app/v4/migration
This PR imports all the work related to Captain into the EE codebase. Captain represents the AI-based features in Chatwoot and includes the following key components:
- Assistant: An assistant has a persona, the product it would be trained on. At the moment, the data at which it is trained is from websites. Future integrations on Notion documents, PDF etc. This PR enables connecting an assistant to an inbox. The assistant would run the conversation every time before transferring it to an agent.
- Copilot for Agents: When an agent is supporting a customer, we will be able to offer additional help to lookup some data or fetch information from integrations etc via copilot.
- Conversation FAQ generator: When a conversation is resolved, the Captain integration would identify questions which were not in the knowledge base.
- CRM memory: Learns from the conversations and identifies important information about the contact.
---------
Co-authored-by: Vishnu Narayanan <vishnu@chatwoot.com>
Co-authored-by: Sojan <sojan@pepalo.com>
Co-authored-by: iamsivin <iamsivin@gmail.com>
Co-authored-by: Sivin Varghese <64252451+iamsivin@users.noreply.github.com>
- This PR adds a UI to validate the response source quality quickly. It also helps to test with sample questions and update responses in the database when missing.
Co-authored-by: Pranav Raj S <pranav@chatwoot.com>
We have been observing JSON parsing errors for responses from GPT. Switching to the gpt-4-1106-preview model along with using response_format has significantly improved the responses from OpenAI, hence making the switch in code.
ref: https://openai.com/blog/new-models-and-developer-products-announced-at-devday
fixes: #CW-2931
This commit introduces the ability to associate response sources to an inbox, allowing external webpages to be parsed by Chatwoot. The parsed data is converted into embeddings for use with GPT models when managing customer queries.
The implementation relies on the `pgvector` extension for PostgreSQL. Database migrations related to this feature are handled separately by `Features::ResponseBotService`. A future update will integrate these migrations into the default rails migrations, once compatibility with Postgres extensions across all self-hosted installation options is confirmed.
Additionally, a new GitHub action has been added to the CI pipeline to ensure the execution of specs related to this feature.
- Initialize an "enterprise" folder that is copyrighted.
- You can remove this folder and the system will continue functioning normally, in case you want a purely MIT licensed product.
- Enable limit on the number of user accounts in enterprise code.
- Use enterprise edition injector methods (inspired from Gitlab).
- SaaS software would run enterprise edition software always.
Co-authored-by: Pranav Raj S <pranav@chatwoot.com>
