recall-growthai

name	recall-growthai
description	Load complete GrowthAI platform context — architecture, setup, deployment, troubleshooting, and support knowledge base
user_invocable	true
trigger	/recall-growthai

/recall-growthai — GrowthAI Support Knowledge Base

When this skill is invoked, load the following context into the conversation and confirm readiness.

---

WHAT IS GROWTHAI

GrowthAI is a multi-tenant SaaS platform by Accelera 360 that automates the full sales lead pipeline via WhatsApp using AI agents (Google Gemini via ADK).

Flow:

Lead enters (Google Forms / CRM / Inbound WhatsApp)
  → Stored in Google Cloud Storage
  → AI Agent contacts lead via WhatsApp (Z-API)
  → Conducts qualifying conversation (BANT or GPCT_LIGHT)
  → Scores lead 0-10 → hot / warm / cold / not_qualified
  → Routes to CRM pipeline (GoHighLevel)
  → Schedules meeting on Google Calendar
  → Sends confirmation email via SMTP

Mission: Accelera 360 accelerates digital transformation in Latin America by implementing AI Agent Flows in client operations.

---

TECH STACK

Layer	Technology
Backend	Python 3.11, Django 5.2, DRF, ASGI (Gunicorn + Uvicorn)
AI/Agents	Google ADK 1.18.0, Gemini 2.5 Flash (fallback: Gemini 2.5 Pro)
Frontend	Next.js 15, React 19, TypeScript 5, Tailwind, shadcn/ui, pnpm
Database	PostgreSQL 15 (prod) / SQLite (dev fallback)
Cache/Sessions	Redis 7 (ADK sessions stored here)
WhatsApp	Z-API (instances per tenant)
CRM	GoHighLevel (GHL) via MCP toolset
Calendar	Google Calendar API (OAuth per tenant)
Storage	Google Cloud Storage (lead files)
Email	Gmail SMTP (app password)
Infra	Docker Compose, GCP VM (e2-medium), NGINX + Certbot SSL
CI/CD	GitLab CI (.gitlab-ci.yml)
Observability	OpenTelemetry + GCP exporters
ML extras	scikit-learn, faiss-cpu, pandas

---

ARCHITECTURE

Docker Services (docker-compose.yml)

postgres-db        ← PostgreSQL 15 (port 5432)
redis-server       ← Redis 7 (internal 6379, external 6380)
migrator           ← One-shot: makemigrations + migrate + bootstrap_app
application-api    ← Django ASGI API (port 8081→8000, WORK=false)
application-api-tasks ← Django background worker (WORK=true, ENABLE_SCHEDULER=true)
application-frontend  ← Next.js (port 3000→80)

Agent Orchestration (Google ADK)

root_agent (gemini-2.5-flash)
│   Tools: consultar_lead, send_whatsapp_message, detect_lead_response,
│          update_context_crm, calendar_tool, lead_locking, atualizar_lead, ghl_mcp
│
├── Score_Classifier    ← BANT/GPCT scoring (0-10)
├── Segregator_Register ← CRM pipeline placement (GHL)
└── SDR_Whats          ← WhatsApp conversation driver

Each call to build_root_agent() creates a fresh isolated agent instance (no race conditions).

Multi-Tenancy

Each user (tenant) has isolated:

• Gemini API key (stored encrypted in DB, resolved by ConfigResolver)
• GHL credentials (location_id, pipeline_id, private_key)
• WhatsApp instance (Z-API instance_id + token)
• Google Calendar OAuth token
• SMTP credentials
• GCS bucket prefix

ConfigResolver priority: TenantSettings (DB) → global defaults → .env (infra fallback only)

Key Django Models (web/models.py)

• CustomUser — AUTH_USER_MODEL, tenant identity
• TenantSettings — 30+ fields, encrypted secrets (Fernet AES-128), auto-created via signal
• WhatsAppMessage — full audit trail per lead per user
• ProcessedLead — idempotency + runner state tracking
• AgentSession — shared session state across processes
• ProfileSettings — per-tenant UI and corporate config
• CalendarOAuthAuditEvent — Google Calendar OAuth audit log
• WebhookEvent — webhook debugging

Request Flow (Webhook → Agent)

WhatsApp reply → POST /webhook/whatsapp/ → web/views.py
  → background task dispatched
  → ADK session resumed from Redis
  → ConfigResolver resolves tenant credentials
  → agent continues conversation

Inbound Unknown Numbers

When a number contacts without existing lead record:

• Tagged with [INBOUND_DESCONHECIDO] in prompt
• SDR switches to welcoming mode (no outbound pitch)
• Distinct from [SESSAO_RETOMADA] (existing lead, new ADK session)

---

KEY DIRECTORIES

app/                        ← Core AI application
  agent.py                  ← Root orchestrator agent factory
  sub_agents/               ← score_classifier/, sdr_whats/, segregator_register/
  tools/                    ← 20+ tool functions (calendar, whatsapp, lead, GHL, scoring, locking)
  jobs/                     ← Background jobs (process_new_leads_whatsapp, update_lead_bucket)
  config/                   ← JSON configs (fallback_rules, lead_mapping, tags)
  gemini_client.py          ← Per-tenant Gemini key resolution
  audio_transcription.py    ← Gemini Files API transcription (audio→text middleware)

web/                        ← Django app
  models.py                 ← All models (CustomUser, TenantSettings, WhatsAppMessage, etc.)
  views/                    ← Views including processing.py (webhook handler, retry logic)
  config_resolver.py        ← Multi-tenant config resolution
  crypto_utils.py           ← Fernet encryption for secrets
  services/                 ← Calendar OAuth, credentials, audit log
  signals.py                ← Auto-create TenantSettings on user creation
  management/commands/       ← whatsapp_session_clear, inspect, history, calendar_oauth_metrics

config/                     ← Django project config
  settings.py               ← All settings including 20+ ADK_* tuning knobs
  urls.py                   ← URL routes
  asgi.py / wsgi.py

frontend/                   ← Next.js 15 app
  pages/                    ← Next.js pages
  components/               ← React components (chat, sidebar, agents UI, settings)
  lib/translations/         ← pt-BR and en-US

policies/                   ← YAML-driven qualification rules
  qualification.yaml        ← BANT/GPCT thresholds (hot>=8, warm>=6, cold>=4)
  cadence.yaml              ← Segment→pipeline mapping (vip>=8.5, priority>=7, nurture>=5)

observability/              ← OpenTelemetry / GCP module
oraicle/                    ← Legacy compatibility shim (oraicle.tools → app.tools)
key/                        ← Google OAuth credentials (gitignored)
auth-calendar-master/       ← Standalone Google Calendar OAuth portal
scripts/                    ← Utility scripts (validar_gemini_keys.py)
tests/                      ← Test suite

---

HOW TO SET UP — LOCAL (Dev)

Prerequisites

• Docker + Docker Compose installed
• .env file configured (see ENV VARS section)
• key/credentials.json (Google OAuth)
• key/mvp-agentes-accelera.json (GCS service account) — optional for local

Steps

# 1. Clone
git clone https://gitlab.com/accelera3601/growth-ai-agents.git
cd growth-ai-agents

# 2. Create .env from template (see README_ENV.md for all vars)
cp .env.example .env  # or create manually

# 3. Start dev environment
docker-compose -f docker-compose.dev.yml up

# This starts:
# - PostgreSQL 15 (port 5432)
# - Redis 7 (port 6380)
# - Backend Django with hot-reload (port 8081)
# - Frontend Next.js dev mode (port 3000)
# - Background tasks worker

# 4. Create superuser
docker exec -it application-api-dev python manage.py createsuperuser

# 5. Access
# Frontend: http://localhost:3000
# Backend API: http://localhost:8081
# Django Admin: http://localhost:8081/admin

Dev Docker Compose differences

• Builds from local Dockerfile (not registry images)
• Volumes mounted for hot-reload (backend: .:/app, frontend: ./frontend:/app)
• Frontend runs pnpm dev instead of production build
• Worker has RUN_DB_SETUP=false to avoid migration race conditions

---

HOW TO DEPLOY — GCP (Production)

Infrastructure Setup (one-time)

1. Create GCP Project — Console → New Project

2. Firewall Rules (5 rules, all Ingress, All instances, 0.0.0.0/0):

   • allow-ssh → tcp:22
   • allow-http → tcp:80
   • allow-https → tcp:443
   • allow-backend-8000 → tcp:8000
   • allow-frontend-3000 → tcp:3000

3. Create VM:

   • Machine: e2-medium (minimum)
   • OS: Ubuntu 22.04 or 24.04 LTS
   • Check: Allow HTTP + HTTPS traffic

4. Static IP:

   • VPC Network → IP Addresses → Reserve Static Address
   • Associate to VM (change from Ephemeral to Static)

5. Service Account (gitlab-ci-deploy):

   • Roles: Compute Instance Admin (v1), Service Account User, Compute Viewer
   • Create JSON key → save as key/gcp-key-work.json

6. Install Docker on VM:

   sudo apt update && sudo apt install -y docker.io git curl wget docker-compose
   sudo systemctl enable docker && sudo systemctl start docker
   sudo usermod -aG docker $USER && newgrp docker
   sudo mkdir -p /opt/app && sudo chown -R $USER:$USER /opt/app
   

CI/CD Pipeline (.gitlab-ci.yml)

Stages: build_frontend → build_backend → deploy

Gate: Only runs on CI_DEPLOY_BRANCH (set in .env)

Deploy process:

1. Authenticates with GCP via service account JSON
2. SCPs docker-compose.yml + .env to VM at /opt/app
3. Pulls new images, runs docker compose up -d --force-recreate
4. Healthcheck: polls 12x (5s each) for all containers healthy/running
5. Auto-rollback on failure: restores docker-compose.rollback.yml

Required CI .env vars:

CI_DEPLOY_BRANCH=feature/ci-cd    # Branch that triggers deploy
CI_REGISTRY_NAMESPACE=accelera3601/growth-ai
CI_FRONTEND_TAG=frontend.v1
CI_BACKEND_TAG=backend.v1
GCP_VM_INSTANCE_NAME=vm-work
GCP_VM_ZONE=us-central1-c
GCP_VM_HOST=<static-ip>
GCP_VM_USER=<vm-username>
GCP_PROJECT_ID=<gcp-project-id>

NGINX + SSL (on VM)

# Install
sudo apt install nginx -y

# Create site config
sudo vim /etc/nginx/sites-available/yourdomain

# Content: proxy / → localhost:3000, /api/ → localhost:8081, /webhooks/ → localhost:8081
# (with long timeouts for webhook/agent processing: 300s read/send timeout)

# Enable + test
sudo ln -s /etc/nginx/sites-available/yourdomain /etc/nginx/sites-enabled/
sudo nginx -t && sudo systemctl reload nginx

# SSL with Certbot
sudo apt install certbot python3-certbot-nginx -y
sudo certbot --nginx -d yourdomain -d www.yourdomain

---

ENV VARS — COMPLETE REFERENCE

Fixed (don't change)

POLLING_INTERVAL_SECONDS=300
EMAIL_HOST=smtp.gmail.com
EMAIL_PORT=587
EMAIL_USE_TLS=True
EMAIL_BACKEND=django.core.mail.backends.smtp.EmailBackend
POSTGRES_HOST=postgres-db       # Docker service name
POSTGRES_PORT=5432
REDIS_HOST=redis
REDIS_PORT=6379
REDIS_DB=0
CI_REGISTRY=registry.gitlab.com
CI_REGISTRY_IMAGE=registry.gitlab.com/accelera360
ADK_APP_NAME=growth-ai-whatsapp
DEPLOY_PATH=/opt/app

Configurable per client/environment

# Django
DJANGO_SECRET_KEY=<generate-unique>
DEBUG=False                      # MUST be False in production
LOG_LEVEL=INFO

# Database
POSTGRES_DB=growthai
POSTGRES_USER=accelera-user
POSTGRES_PASSWORD=<secure-password>

# Encryption
SETTINGS_CRYPTO_KEY=<fernet-key>  # python -c 'from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())'

# Frontend
FRONTEND_DOMAIN=https://yourdomain.com
NEXT_PUBLIC_API_URL=https://yourdomain.com   # In docker-compose.yml line ~107

# Email/SMTP
EMAIL_HOST_USER=[e-mail]
EMAIL_HOST_PASSWORD=<gmail-app-password>     # No spaces!
DEFAULT_FROM_EMAIL=[e-mail]

# Google Calendar
GOOGLE_CALENDAR_ID=primary
GOOGLE_CREDENTIALS_PATH=key/credentials.json
TIMEZONE=America/Sao_Paulo

# Google Cloud Storage
USE_GCS_LEADS=true
GCS_BUCKET_NAME=lead_contacts
GCS_PROJECT_ID=<gcp-project-id>
GOOGLE_APPLICATION_CREDENTIALS=key/mvp-agentes-accelera.json

# WhatsApp (Z-API) — per tenant
Z_API_INSTANCE_ID=<instance>
Z_API_TOKEN=<token>
Z_API_CLIENT_TOKEN=<client-token>

# GoHighLevel — per tenant
USE_GHL_API=true
GHL_MCP_URL=https://services.leadconnectorhq.com/mcp/
GHL_LOCATION_ID=<location>
GHL_CALENDAR_ID=<calendar>
GHL_PIPELINE_ID=<pipeline>
GHL_PIT=<pit>
GHL_PRIVATE_KEY=<private-key>

# Business
ESTADO_ALUNO=SC                  # State for holiday rules (2-letter code)
BUSINESS_HOURS_START=9
BUSINESS_HOURS_END=18
FERIADO_ESTADUAL=True

# Branding
TITULO_MENU="Accelera"
NAME_ATENDENTE_CHAT="Kelvin"
CORPORATE_NAME="ACCELERA360"
LINK_PROJETO="https://accelera360.com.br/"

ADK Tuning Knobs (advanced)

ADK_MAX_CONCURRENT_CALLS=50
ADK_MIN_REQUEST_INTERVAL_SECONDS=0.5
ADK_AGENT_EXECUTION_TIMEOUT_SECONDS=180
ADK_503_MAX_ATTEMPTS=5
ADK_503_RETRY_DELAY_SECONDS=3.0
ADK_503_BACKOFF_MULT=2.0
ADK_503_MAX_DELAY_SECONDS=30.0
ADK_503_JITTER_SECONDS=1.0
ADK_503_FALLBACK_MODEL=gemini-2.5-pro
ADK_RESOURCE_EXHAUSTED_MAX_ATTEMPTS=4
ADK_RESOURCE_EXHAUSTED_RETRY_DELAY_SECONDS=2.0
ADK_RESOURCE_EXHAUSTED_BACKOFF_MULT=1.5
ADK_RESOURCE_EXHAUSTED_MAX_DELAY_SECONDS=30.0
ADK_APPOINTMENT_EARLY_FINALIZE_SECONDS=12

---

GOOGLE SETUP (Calendar + SMTP)

Google Cloud Console (https://console.cloud.google.com/)

1. Create/select project
2. Enable APIs: Google Calendar API + Google Drive API
3. OAuth Consent Screen: External, add company email as test user
4. Create OAuth credentials: Desktop app → download credentials.json → save to key/credentials.json
5. First run: browser opens, login with test user, authorize → creates key/token.json

Gmail SMTP

1. Enable 2FA: https://myaccount.google.com/security
2. Create App Password: https://myaccount.google.com/apppasswords → "Growth AI"
3. Use generated password (remove spaces) in EMAIL_HOST_PASSWORD

Per-Tenant Gemini Keys (GA-007)

• Each tenant MUST have their own Gemini API key in profile settings
• Without it: GEMINI_NOT_CONFIGURED / SetupPendingError
• Get key: https://aistudio.google.com/app/apikey
• Validate: python scripts/validar_gemini_keys.py
• Logs: source=tenant = correct; source=env = missing key (red flag in prod)

---

TROUBLESHOOTING

Agent not responding

1. Check container is running: docker ps
2. Check logs: docker compose logs -f application-api
3. Check Redis: docker exec redis-server redis-cli ping
4. Check Gemini key: GET /api/v1/tenant/settings → google_api_key_configured: true?
5. Check ADK session: python manage.py whatsapp_session_inspect --phone <number> --user-id <id>

Webhook not receiving

1. Check NGINX config: sudo nginx -t
2. Check webhook URL in Z-API points to https://yourdomain/webhook/whatsapp/
3. Check SSL: sudo certbot certificates
4. Check container: docker logs application-api -f --tail=100
5. Check WebhookEvent table in Django admin

Agent repeats greeting / loses context

1. Session may have been recreated: check for [SESSAO_RETOMADA] in logs
2. Redis may have evicted session: check Redis memory (docker exec redis-server redis-cli info memory)
3. Clear and restart: python manage.py whatsapp_session_clear --phone <number> --user-id <id>

503 errors / Gemini overloaded

1. Check logs for [503_RETRY], [503_FALLBACK], [503_FINAL]
2. Tune: ADK_503_MAX_ATTEMPTS, ADK_503_RETRY_DELAY_SECONDS
3. Fallback model: ADK_503_FALLBACK_MODEL=gemini-2.5-pro
4. Consider: user might need higher Vertex quota

Calendar not working

1. Check OAuth: GET /api/calendar/oauth/status
2. Check credentials: key/credentials.json exists, key/token.json generated
3. Check logs: filter by [CALENDAR_OAUTH]
4. Metrics: python manage.py calendar_oauth_metrics --days 7
5. invalid_grant = token expired, need re-auth

Deploy failed

1. Check GitLab CI pipeline logs
2. Verify .env CI vars (CI_DEPLOY_BRANCH matches pushed branch)
3. Check key/gcp-key-work.json exists and SA has correct roles
4. SSH to VM: gcloud compute ssh <user>@<vm> --zone <zone>
5. On VM: docker compose logs, docker ps -a

Database issues

1. Check migrations: docker exec application-api python manage.py showmigrations
2. Force migrate: docker exec application-api python manage.py migrate
3. Check connection: docker exec postgres-db pg_isready

---

MANAGEMENT COMMANDS

# WhatsApp session management
python manage.py whatsapp_session_clear --phone <number> --user-id <id>
python manage.py whatsapp_session_inspect --phone <number> --user-id <id>
python manage.py whatsapp_session_history --phone <number> --user-id <id> --limit 20

# Calendar OAuth metrics
python manage.py calendar_oauth_metrics --days 7

# Validate Gemini keys across tenants
python scripts/validar_gemini_keys.py

# Create superuser
python manage.py createsuperuser

# Bootstrap app (creates default data)
python manage.py bootstrap_app

# Generate encryption key
python -c 'from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())'

# Test calendar connection
python -c "from app.tools.calendar_tool import check_calendar_availability; print(check_calendar_availability())"

---

QUALIFICATION & CADENCE

Scoring Frameworks

• BANT: Budget (25pts) + Authority (25pts) + Need (25pts) + Timeline (25pts) = 100 → normalized to 0-10
• GPCT_LIGHT (simplified): Interesse (50pts) + Capacidade (50pts) = 100 → normalized to 0-10

Thresholds

Score	Label	Action
>= 8.0	hot	Immediate handoff
6.0-7.9	warm	Active follow-up
4.0-5.9	cold	Controlled nurturing
< 4.0	not_qualified	Long-term

Cadence Mapping

Score	Segment	Pipeline	Frequency
8.5-10.0	vip	vip_pipeline	Daily
7.0-8.4	priority	priority_pipeline	2x/week
5.0-6.9	nurture	nurture_program	Weekly
0-4.9	long_term	segmentation_pending	Monthly

---

API ROUTES (key endpoints)

# Auth
POST /api/auth/register
POST /api/auth/login
POST /api/auth/logout
POST /api/auth/forgot-password
POST /api/auth/reset-password

# Profile & Settings
GET|PATCH /api/v1/tenant/settings        ← tenant config (masked secrets)
GET /api/v1/tenant/health                ← per-module health status
GET|PATCH /api/profile/settings

# Calendar OAuth
GET /api/calendar/oauth/start            ← initiates OAuth flow
GET /api/calendar/oauth/callback         ← OAuth redirect
GET /api/calendar/oauth/status           ← connection status
POST /api/calendar/oauth/disconnect

# WhatsApp
GET /api/whatsapp/status
POST /api/whatsapp/connect
POST /api/whatsapp/disconnect

# Agents
GET /api/agents/status
GET /api/agents/metrics
GET /api/agents/sessions
GET /api/agents/conversations/active

# Webhooks
POST /webhook/whatsapp/                  ← Z-API webhook (multiple alias paths)

# Metrics
GET /api/metricas/

---

RELEASE NOTES (Latest: 2026-03-27)

Key features in current release:

• Audio input support — agent receives audio + text, always responds in text
• Inbound unknown numbers — [INBOUND_DESCONHECIDO] tag for first-contact leads not in DB
• GHL + Google Calendar integration for automatic scheduling
• Per-tenant Gemini keys (GA-007) — each client uses their own API key
• 503 retry with fallback — exponential backoff + model fallback on Gemini overload
• Frontend-driven config — Calendar, SMTP settings all via UI, not .env

---

TRAINING VIDEOS (from Accelera Drive)

Videos are stored on Google Drive. Each video has a dedicated .md file in videos/ with full timestamps, commands, checklists, and troubleshooting.

Video Index

#	Titulo	Duracao	Categoria	Arquivo
01	Clonando o Repositorio	~2:33	Setup Inicial	videos/01-clonando-repositorio.md
02	Arquivo .env	~0:19	Setup Inicial	videos/02-arquivo-env.md
03	Missao e Identidade	~0:58	Branding	videos/03-missao-identidade.md
04	Frontend e Variaveis	~2:30	Configuracao	videos/04-frontend-variaveis.md
05	Rodando Local (Docker + Ngrok)	~3:26	Setup Local	videos/05-rodando-local-docker-ngrok.md
06	Configuracao do GHL	~2:47	Integracao CRM	videos/06-configuracao-ghl.md
07	Google AI Studio (Gemini Key)	~1:23	Integracao AI	videos/07-google-ai-studio-gemini.md
08	Configuracao da Z-API	~1:19	Integracao WhatsApp	videos/08-configuracao-zapi.md
09	GCP: Firewall + VM + Service Account	~6:44	Infraestrutura GCP	videos/09-gcp-firewall-vm-serviceaccount.md
10	IP Estatico na VM	~1:45	Infraestrutura GCP	videos/10-ip-estatico-gcp.md

Quick Reference — Common Support Questions → Video

Pergunta do cliente	Video	Timestamp
"Como clono o repositorio?"	01	0:40
"Qual meu username do GitLab?"	01	1:00
"Como configuro o remote?"	01	2:00
"Cadê o arquivo .env?"	02	0:10
"Como renomear env.example?"	02	0:12
"Como mudo o nome do bot no WhatsApp?"	03	0:39 (NAME_ATENDENTE_CHAT)
"Como mudo o nome da empresa?"	03	0:41 (CORPORATE_NAME)
"Onde configuro o dominio?"	04	0:28 (api.ts), 0:55 (.env), 1:18 (docker-compose)
"Como crio o superusuario?"	04	1:54
"Frontend nao conecta na API"	04	Checar os 3 arquivos de dominio
"Como rodo local com Docker?"	05	0:39 (docker build) → 3:06 (docker-compose up)
"Como configuro o Ngrok?"	05	1:28 (signup) → 2:43 (ngrok http 8081)
"Onde coloco o link do Ngrok?"	05	2:52 (ALLOWED_HOSTS + WEBHOOK_URL)
"Como pego o Location ID do GHL?"	06	0:33 (Configuracoes > Perfil)
"Como crio o PIT no GHL?"	06	0:53 (Integracoes > Privado > Criar)
"Como pego o Pipeline ID?"	06	1:40 (Leads > Pipelines > Copiar link > extrair ID)
"Como pego o Calendar ID do GHL?"	06	2:24 (Calendarios > Configuracoes)
"Como crio a API Key do Gemini?"	07	0:42 (Get API key > Create)
"Onde configuro a Z-API?"	08	0:05 (Client Token) → 0:22 (Instance Token) → 0:39 (Instance ID)
"Como configuro o webhook na Z-API?"	08	0:54 (Webhook > Ao receber > colar URL)
"Como crio as regras de firewall no GCP?"	09	0:16 (Firewall > Create) — 5 regras
"Como crio a VM no GCP?"	09	3:02 (Compute Engine > Create Instance)

Erros Mais Comuns por Video

Video	Erro	Causa raiz
01	"Authentication failed"	2FA ativado no GitLab, precisa de Personal Access Token
01	Push vai pro repo errado	Esqueceu git remote remove origin
02	App nao inicia	Esqueceu de renomear env.example → .env
03	Bot se apresenta como "Accelera"	NAME_ATENDENTE_CHAT / CORPORATE_NAME nao alterados
04	ERR_CONNECTION_REFUSED em producao	Dominio alterado em 1 ou 2 dos 3 lugares, mas nao nos 3
04	Login nao funciona	APP_USERNAME/APP_PASSWORD vazios no .env
05	"image not found" no docker-compose	Tags no .env nao batem com nomes do docker build
05	"DisallowedHost" no Django	Hostname do ngrok nao esta no ALLOWED_HOSTS
06	"GHL connection failed"	PIT expirado ou escopos incompletos — recriar integracao
06	Leads nao aparecem no pipeline	GHL_PIPELINE_ID errado — extrair so o ID da URL
07	GEMINI_NOT_CONFIGURED	GOOGLE_API_KEY vazia ou errada no .env
07	"Permission denied" 403	"Generative Language API" nao habilitada no projeto GCP
08	Mensagens enviadas mas respostas nao chegam	Webhook nao configurado no painel Z-API
08	Funciona local mas para apos reiniciar	URL do Ngrok mudou — atualizar .env + painel Z-API
09	"Quota exceeded" ao criar VM	Cota de CPUs esgotada — trocar zona ou pedir aumento
09	SSH nao conecta na VM	Regra allow-ssh nao criada — verificar firewall
09

Cada arquivo de video contem:

• Resumo do que o video ensina
• Timeline detalhada segundo a segundo
• Comandos exatos executados (copiaveis)
• Checklist de verificacao pos-video
• Tabela de problemas comuns + solucoes
• Notas tecnicas para suporte avancado
• Links para documentacao relacionada

Para suporte: identifique o problema → encontre o video na tabela acima → abra o .md → va direto ao timestamp + solucao.

---

SUPPORT WORKFLOW

When providing technical support to Accelera clients:

1. Identify the issue category — setup, deployment, runtime, integration
2. Check the relevant section above
3. Use management commands to diagnose (session inspect, metrics, validate keys)
4. Check logs with appropriate prefixes: [GEMINI], [CALENDAR_OAUTH], [503_RETRY], [INBOUND_DESCONHECIDO]
5. Reference training videos from Drive for step-by-step visual guidance
6. Escalate if: database corruption, security incident, or Gemini API quota/billing issues

---

RAG — BUSCA SEMANTICA AUTOMATICA

INSTRUCAO CRITICA PARA O CLAUDE: Quando esta skill estiver carregada e o usuario fizer QUALQUER pergunta sobre GrowthAI, SEMPRE execute este comando Bash ANTES de responder:

PYTHONUNBUFFERED=1 python C:/Users/Pichau/.claude/skills/recall-growthai/rag/query_knowledge.py "PERGUNTA_AQUI"

Substitua PERGUNTA_AQUI pela pergunta do usuario (em portugues ou ingles). Use timeout de 120000ms (2 min) pois o modelo de embeddings demora ~90s no primeiro load da sessao. Depois fica rapido. Use PYTHONUNBUFFERED=1 antes do python para garantir output.

Base: 1,110 chunks de 61 arquivos (21 videos + repo docs + .env.example + policies + release notes).

Regras:

1. SEMPRE rode a query antes de responder perguntas tecnicas sobre GrowthAI
2. Use os resultados HIGH como base da resposta
3. Cite o source (video + timestamp ou doc) na resposta
4. Se nenhum resultado e HIGH, diga que nao tem certeza e sugira onde procurar
5. Se o comando demorar, nao desista — o primeiro load do modelo leva ~15s, depois fica rapido
6. Para re-indexar apos novos videos: python C:/Users/Pichau/.claude/skills/recall-growthai/rag/index_knowledge.py

---

Context loaded. Ready for GrowthAI support.