Memória de Longo Prazo para IA: Como o Graphiti Funciona para Construir Aplicações Realmente Inteligentes

O Graphiti transforma dados em constante mudança em um grafo de conhecimento bi-temporal que os seus agentes de IA podem consultar em milissegundos.

1 O que é o Graphiti?

Graphiti é um framework Python open-source para construir grafos de conhecimento em tempo real e bi-temporais, projetados para geração aumentada por recuperação (RAG) e agentes autônomos. Ele ingere continuamente eventos estruturados, semiestruturados e não estruturados — como conversas, documentos e transações — e anexa metadados temporais para que você sempre saiba quando algo aconteceu e quando o sistema aprendeu sobre isso.

Capacidades principais

A imagem do MCP Server empacota o Graphiti como um container com interfaces REST e WebSocket. Aponte para Neo4j >= 5.26, exporte OPENAI_API_KEY, e você estará no ar em minutos.

2 Instalando o Graphiti em 5 Minutos

Siga estes passos em qualquer laptop com Docker:

1. Suba o Graphiti MCP Server

   docker run -d --name graphiti \
     -p 8000:8000 \
     -e OPENAI_API_KEY=$OPENAI_API_KEY \
     zepai/graphiti:latest

2. Envie seu primeiro evento via POST

   curl -X POST http://localhost:8000/v1/episodes \
     -H "Content-Type: application/json" \
     -d '{
       "actor":"user:123",
       "content":"Paid $45 at Starbucks",
       "timestamp":"2025-07-14T15:04:22Z"
     }'

Você acabou de criar um Episode com tag temporal que a IA agora consegue entender e consultar.

3 Como o Graphiti Transforma Dados Brutos em Memória Inteligente para IA

graph LR
A["Vida Diária de Sarah<br/>(transações, contas, renda)"] -->|"Graphiti Aprende"| B["Sistema de Memória Inteligente"]
B --> C["Conhecimento Organizado<br/>Sobre Sarah"]
C --> D["IA Entende<br/>Seus Padrões"]
C --> E["IA Reconhece<br/>Seus Hábitos"]
D --> F["ChatGPT Dá<br/>Conselhos Pessoais"]
E --> F

O Graphiti é a memória financeira da Sarah que nunca esquece nada:

• Quando Sarah gasta $45 no Starbucks, o Graphiti entende "Sarah comprou café no Starbucks por $45."
• Ele lembra não apenas o que aconteceu, mas quando aconteceu e quando o sistema soube disso.
• Se Sarah começa a comprar café diariamente após se mudar para o centro, o Graphiti conecta esses eventos.
• Quando Sarah pergunta "Quanto gasto com café?", o Graphiti recupera todas as informações relacionadas instantaneamente.
• O ChatGPT recebe exatamente o contexto certo sobre os hábitos de café da Sarah para dar conselhos úteis.

4 Casos de Uso Além do Histórico de Chat

5 Construindo um Coach Financeiro Pessoal com IA

Vamos criar um sistema que ingere transações diárias, aprende padrões de consumo e fornece coaching financeiro personalizado.

5.1 Como Hábitos Financeiros Viram Conhecimento em Grafo

Aqui está a jornada financeira de Sarah, mostrando como o Graphiti transforma suas atividades diárias em insights acionáveis.

Eventos Financeiros Brutos

O comportamento financeiro real de Sarah é capturado ao longo do seu dia:

// Morning coffee routine - Episode 1
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $45 at Starbucks on Main Street for a venti caramel macchiato and pastry",
  "timestamp": "2025-07-14T08:15:00Z",
  "metadata": {
    "amount": 45.00,
    "merchant": "Starbucks",
    "category": "coffee",
    "location": "Main Street",
    "items": ["venti caramel macchiato", "blueberry muffin"],
    "frequency": "daily"
  }
}

// Grocery shopping - Episode 2
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $127.50 at Whole Foods for weekly groceries",
  "timestamp": "2025-07-14T10:30:00Z",
  "metadata": {
    "amount": 127.50,
    "merchant": "Whole Foods",
    "category": "groceries",
    "frequency": "weekly",
    "items": ["organic vegetables", "premium meat", "specialty cheese"]
  }
}

// Gas station visit - Episode 3
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $52.00 on gas at Shell station",
  "timestamp": "2025-07-14T12:45:00Z",
  "metadata": {
    "amount": 52.00,
    "merchant": "Shell",
    "category": "transportation",
    "frequency": "bi-weekly",
    "gallons": 12.5
  }
}

// Lunch with colleague - Episode 4
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $38.50 at Olive Garden for lunch meeting",
  "timestamp": "2025-07-14T13:20:00Z",
  "metadata": {
    "amount": 38.50,
    "merchant": "Olive Garden",
    "category": "dining",
    "occasion": "business lunch",
    "frequency": "occasional"
  }
}

// Impulse purchase - Episode 5
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $85.00 at Target for household items and unplanned clothing",
  "timestamp": "2025-07-14T16:00:00Z",
  "metadata": {
    "amount": 85.00,
    "merchant": "Target",
    "category": "retail",
    "planned_amount": 25.00,
    "impulse_amount": 60.00,
    "items": ["laundry detergent", "sweater", "decorative pillows"]
  }
}

// Evening entertainment - Episode 6
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $67.00 at AMC Theater for movie and snacks",
  "timestamp": "2025-07-14T19:30:00Z",
  "metadata": {
    "amount": 67.00,
    "merchant": "AMC Theater",
    "category": "entertainment",
    "frequency": "weekly",
    "companions": 2
  }
}

// Online shopping - Episode 7
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $142.00 on Amazon for books and electronics",
  "timestamp": "2025-07-14T21:15:00Z",
  "metadata": {
    "amount": 142.00,
    "merchant": "Amazon",
    "category": "online_shopping",
    "items": ["wireless headphones", "self-help books", "phone charger"],
    "delivery": "next_day"
  }
}

// Late night food delivery - Episode 8
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $28.50 on DoorDash for Thai food delivery",
  "timestamp": "2025-07-14T22:45:00Z",
  "metadata": {
    "amount": 28.50,
    "merchant": "DoorDash",
    "category": "food_delivery",
    "restaurant": "Thai Garden",
    "frequency": "3x_weekly"
  }
}

// Bill payment delay - Episode 9
{
  "actor": "user:sarah_jones", 
  "content": "Sarah's credit card payment of $1,200 is 5 days overdue",
  "timestamp": "2025-07-14T23:59:00Z",
  "metadata": {
    "amount": 1200.00,
    "payment_type": "credit_card",
    "days_overdue": 5,
    "severity": "moderate",
    "min_payment": 35.00
  }
}

// Morning workout - Episode 10
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $89.00 for monthly gym membership at Planet Fitness",
  "timestamp": "2025-07-15T07:00:00Z",
  "metadata": {
    "amount": 89.00,
    "merchant": "Planet Fitness",
    "category": "fitness",
    "frequency": "monthly",
    "auto_pay": true
  }
}

// Utility bill - Episode 11
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $156.00 for electricity bill to ConEd",
  "timestamp": "2025-07-15T09:30:00Z",
  "metadata": {
    "amount": 156.00,
    "merchant": "ConEd",
    "category": "utilities",
    "frequency": "monthly",
    "kwh_used": 890,
    "season": "summer"
  }
}

// Subscription service - Episode 12
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $15.99 for Netflix subscription",
  "timestamp": "2025-07-15T10:45:00Z",
  "metadata": {
    "amount": 15.99,
    "merchant": "Netflix",
    "category": "entertainment",
    "frequency": "monthly",
    "auto_pay": true
  }
}

// Healthcare expense - Episode 13
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $35.00 copay for doctor visit",
  "timestamp": "2025-07-15T11:15:00Z",
  "metadata": {
    "amount": 35.00,
    "merchant": "Dr. Smith Medical",
    "category": "healthcare",
    "frequency": "quarterly",
    "insurance_covered": 85.00
  }
}

// Pharmacy visit - Episode 14
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $42.50 at CVS for prescription and vitamins",
  "timestamp": "2025-07-15T12:00:00Z",
  "metadata": {
    "amount": 42.50,
    "merchant": "CVS",
    "category": "healthcare",
    "items": ["prescription", "multivitamins", "face wash"],
    "insurance_covered": 15.00
  }
}

// Car maintenance - Episode 15
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $1,200 for car repair at AutoFix",
  "timestamp": "2025-07-15T14:45:00Z",
  "metadata": {
    "amount": 1200.00,
    "merchant": "AutoFix",
    "category": "emergency",
    "payment_method": "credit_card",
    "urgency": "high",
    "repair_type": "transmission"
  }
}

// Side income - Episode 16
{
  "actor": "user:sarah_jones",
  "content": "Sarah received $350 from freelance graphic design work",
  "timestamp": "2025-07-15T16:20:00Z",
  "metadata": {
    "amount": 350.00,
    "source": "freelance",
    "category": "additional_income",
    "frequency": "irregular",
    "client": "Small Business Co",
    "hours_worked": 8
  }
}

// Insurance payment - Episode 17
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $245.00 for auto insurance to State Farm",
  "timestamp": "2025-07-15T17:00:00Z",
  "metadata": {
    "amount": 245.00,
    "merchant": "State Farm",
    "category": "insurance",
    "frequency": "monthly",
    "coverage": "full",
    "auto_pay": true
  }
}

// Weekend shopping - Episode 18
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $180.00 at Zara for new work clothes",
  "timestamp": "2025-07-15T18:30:00Z",
  "metadata": {
    "amount": 180.00,
    "merchant": "Zara",
    "category": "clothing",
    "frequency": "monthly",
    "items": ["blazer", "blouse", "dress pants"],
    "occasion": "work_wardrobe"
  }
}

// Pet expenses - Episode 19
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $95.00 at PetSmart for dog food and toys",
  "timestamp": "2025-07-15T19:15:00Z",
  "metadata": {
    "amount": 95.00,
    "merchant": "PetSmart",
    "category": "pet_care",
    "frequency": "monthly",
    "items": ["premium dog food", "chew toys", "treats"],
    "pet": "golden_retriever"
  }
}

// Investment contribution - Episode 20
{
  "actor": "user:sarah_jones",
  "content": "Sarah invested $500 in index funds through her 401k",
  "timestamp": "2025-07-15T20:00:00Z",
  "metadata": {
    "amount": 500.00,
    "investment_type": "401k",
    "asset_class": "index_funds",
    "frequency": "monthly",
    "employer_match": 250.00
  }
}

// Rent payment - Episode 21
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $1,800 rent to Sunset Apartments",
  "timestamp": "2025-07-16T08:00:00Z",
  "metadata": {
    "amount": 1800.00,
    "merchant": "Sunset Apartments",
    "category": "housing",
    "frequency": "monthly",
    "payment_method": "auto_pay",
    "apartment": "2BR_downtown"
  }
}

// Phone bill - Episode 22
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $78.00 for Verizon phone service",
  "timestamp": "2025-07-16T09:30:00Z",
  "metadata": {
    "amount": 78.00,
    "merchant": "Verizon",
    "category": "utilities",
    "frequency": "monthly",
    "plan": "unlimited",
    "auto_pay": true
  }
}

// Loan application - Episode 23
{
  "actor": "user:sarah_jones",
  "content": "Sarah applied for a $15,000 personal loan for debt consolidation",
  "timestamp": "2025-07-16T14:30:00Z", 
  "metadata": {
    "amount": 15000.00,
    "loan_type": "personal",
    "purpose": "debt_consolidation",
    "interest_rate": 8.5,
    "term_months": 36
  }
}

// Subscription cancellation - Episode 24
{
  "actor": "user:sarah_jones",
  "content": "Sarah canceled her $19.99 Spotify Premium subscription",
  "timestamp": "2025-07-16T15:45:00Z",
  "metadata": {
    "amount": 19.99,
    "merchant": "Spotify",
    "category": "entertainment",
    "action": "cancel",
    "reason": "cost_cutting"
  }
}

// Weekend dining - Episode 25
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $125.00 at The Cheesecake Factory for dinner with friends",
  "timestamp": "2025-07-16T19:00:00Z",
  "metadata": {
    "amount": 125.00,
    "merchant": "The Cheesecake Factory",
    "category": "dining",
    "frequency": "weekly",
    "companions": 3,
    "occasion": "social"
  }
}

// Online course - Episode 26
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $299.00 on Udemy for Python programming course",
  "timestamp": "2025-07-17T10:00:00Z",
  "metadata": {
    "amount": 299.00,
    "merchant": "Udemy",
    "category": "education",
    "course": "Python programming",
    "duration": "40_hours",
    "career_development": true
  }
}

// Emergency fund withdrawal - Episode 27
{
  "actor": "user:sarah_jones",
  "content": "Sarah withdrew $800 from savings account for emergency fund",
  "timestamp": "2025-07-17T11:30:00Z",
  "metadata": {
    "amount": 800.00,
    "account_type": "savings",
    "purpose": "emergency_fund",
    "remaining_balance": 2400.00,
    "reason": "car_repair_payment"
  }
}

// Birthday gift - Episode 28
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $75.00 at Bath & Body Works for friend's birthday gift",
  "timestamp": "2025-07-17T13:15:00Z",
  "metadata": {
    "amount": 75.00,
    "merchant": "Bath & Body Works",
    "category": "gifts",
    "frequency": "occasional",
    "occasion": "birthday",
    "recipient": "friend"
  }
}

// Meal prep ingredients - Episode 29
{
  "actor": "user:sarah_jones",
  "content": "Sarah spent $85.00 at Costco for meal prep ingredients",
  "timestamp": "2025-07-17T15:00:00Z",
  "metadata": {
    "amount": 85.00,
    "merchant": "Costco",
    "category": "groceries",
    "purpose": "meal_prep",
    "frequency": "bi-weekly",
    "bulk_purchase": true
  }
}

// Internet bill - Episode 30
{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $89.99 for Comcast internet service",
  "timestamp": "2025-07-17T16:30:00Z",
  "metadata": {
    "amount": 89.99,
    "merchant": "Comcast",
    "category": "utilities",
    "frequency": "monthly",
    "speed": "100mbps",
    "auto_pay": true
  }
}

Como a IA Aprende os Padrões Financeiros de Sarah

Assim que as atividades de Sarah estão no Graphiti, a IA consegue raciocinar sobre elas como um consultor financeiro que acompanha sua conta há meses:

1. Ela identifica entidades: Sarah, Starbucks, Target, seu cartão de crédito, seu fundo de emergência, sua renda freelance.
2. Ela conecta padrões: "Sarah vai ao Starbucks todo dia às 8h15 por $45" constrói reconhecimento de hábitos. "Sarah paga $1.800 de aluguel + $156 de eletricidade + $89 de academia + $78 de telefone" rastreia custos fixos. "Sarah cancelou o Spotify por $19,99" sinaliza intenção de cortar custos.
3. Ela entende o perfil financeiro dela: café + delivery de comida = $1.440/mês em gastos discricionários com alimentação. Aluguel + utilidades + seguro + academia = $2.634/mês em custos fixos. Pagamentos atrasados, saques do fundo de emergência e uma solicitação de empréstimo sinalizam estresse financeiro.
4. Ela rastreia mudanças ao longo do tempo: os gastos subiram 15% após o conserto do carro. Os atrasos nos pagamentos começaram 3 meses atrás. O corte de custos começou 2 semanas atrás.

Por que Graphiti ao invés de Bancos de Dados SQL Tradicionais?

O problema do SQL para esse tipo de análise:

-- Traditional SQL struggles with temporal relationships
SELECT u.name, t.amount, t.category, t.timestamp
FROM users u JOIN transactions t ON u.id = t.user_id
WHERE u.id = 'sarah_jones' AND t.timestamp > '2025-07-01'
ORDER BY t.timestamp;

Isso te dá dados brutos, mas não consegue explicar padrões comportamentais, contexto temporal sobre quando os hábitos mudaram, relacionamentos entre eventos, ou por que os gastos aumentaram.

A abordagem do Graphiti:

# Graphiti transforms raw events into AI-queryable knowledge
coffee_habits = await graphiti.search(
    "Sarah coffee spending habits frequency daily Starbucks",
    limit=20
)

# AI can now understand:
# - Sarah visits Starbucks every morning at 8:15 AM
# - She spends $45 per visit (above average for coffee)
# - This habit represents 12% of her monthly income
# - It started 6 months ago when she moved downtown
# - It correlates with her recent payment delays

Comparação:

O contraste fica mais claro com um exemplo específico.

SQL:

SELECT AVG(amount) as avg_coffee_spend, COUNT(*) as frequency
FROM transactions 
WHERE user_id = 'sarah_jones' 
AND merchant LIKE '%Starbucks%' 
AND timestamp >= DATE_SUB(NOW(), INTERVAL 30 DAY);
-- Result: $45 average, 22 visits
-- But what does this mean for Sarah?

Graphiti:

# Search for coffee patterns
coffee_data = await graphiti.search(
    "Sarah coffee spending habits daily routine financial impact",
    limit=15
)

# AI analyzes and provides personalized insight
insight = await chatgpt.analyze(f"""
Sarah's coffee data: {coffee_data}
Provide personalized financial advice for Sarah about her coffee spending.
""")

# Result: "Sarah, your daily $45 Starbucks visits cost $990/month - 
# that's 12% of your income! Try brewing premium coffee at home 
# for $8/month. You could save $982 monthly for your emergency fund."

O SQL diz que ela gastou $45 no Starbucks. O Graphiti diz que ela iniciou esse hábito 6 meses atrás quando se mudou para perto do escritório no centro, que isso se correlaciona com o aumento do estresse no trabalho, e que coincide com os atrasos recentes nos pagamentos.

5.2 Como o Sistema Funciona

1. Os dados de Sarah vêm de transações de cartão de crédito e banco, atividade em aplicativos financeiros e histórico de pagamentos.
2. Um coletor de dados transforma cada ação financeira em uma narrativa que o Graphiti pode armazenar e pesquisar.
3. O Graphiti salva tudo de forma que a IA possa consultar eficientemente.
4. O ChatGPT lê os padrões de Sarah e gera uma pontuação de saúde financeira de 0 a 100.
5. Um gerador de insights pessoais analisa seus últimos 90 dias e produz conselhos específicos e acionáveis.
6. Um dashboard mostra a Sarah seus insights e recomendações gerados pela IA.

5.2 O que a IA Busca nos Hábitos Financeiros de Sarah

5.3 Insights Financeiros Pessoais Gerados para Sarah

Todos os insights são gerados primeiro usando a funcionalidade de busca do Graphiti para recuperar informações relevantes, e então alimentando esses resultados ao ChatGPT para conselhos personalizados.

Otimização dos Gastos com Café

Passo 1: Busca pelos Dados de Gastos com Café

# Search for coffee-related spending patterns
coffee_search_results = await graphiti.search(
    "Sarah coffee spending Starbucks habits frequency amount daily routine",
    limit=15
)

# Search for income and budget context
income_search_results = await graphiti.search(
    "Sarah monthly income salary freelance total earnings",
    limit=10
)

Passo 2: Conselho Personalizado da IA para Sarah

# Extract facts from search results
coffee_facts = [result.fact for result in coffee_search_results]
income_facts = [result.fact for result in income_search_results]

# Create personalized advice prompt
ai_prompt = f"""
You are Sarah's personal financial AI assistant. Based on her coffee spending data, provide friendly, actionable advice.

Sarah's Coffee Spending:
{coffee_facts}

Sarah's Income Context:
{income_facts}

Provide personalized advice to Sarah:
1. Friendly observation about her coffee habits
2. Financial impact in terms she can understand
3. Specific, actionable alternatives
4. Monthly savings potential
5. Motivation and encouragement
"""

personal_insight = await client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": ai_prompt}]
)

Insight Pessoal Gerado:

"Oi Sarah! Percebi que você está gastando $45 por dia no Starbucks — isso dá $990 por mês, ou seja, 12% da sua renda. Entendo completamente — um bom café torna o dia melhor! Mas aqui está uma ideia: e se você investisse em uma máquina de espresso premium por $300? Você poderia fazer bebidas com qualidade de café em casa por cerca de $8/mês. Isso representa uma economia de $982 mensais, o suficiente para reconstituir seu fundo de emergência em apenas 3 meses! Além disso, você ganharia mais tempo de manhã. Quer que eu encontre algumas máquinas de espresso bem avaliadas dentro do seu orçamento?"

Lembrete de Pagamento e Gestão de Fluxo de Caixa

Passo 1: Busca pelos Padrões de Pagamento

# Search for payment history and upcoming bills
payment_search_results = await graphiti.search(
    "Sarah payment history missed late overdue bills upcoming",
    limit=20
)

# Search for cash flow patterns
cashflow_search_results = await graphiti.search(
    "Sarah income expenses monthly flow timing emergency fund",
    limit=15
)

Passo 2: Assistência Proativa ao Pagamento

payment_facts = [result.fact for result in payment_search_results]
cashflow_facts = [result.fact for result in cashflow_search_results]

ai_prompt = f"""
You are Sarah's helpful financial assistant. Based on her payment history, provide supportive guidance.

Payment History:
{payment_facts}

Cash Flow Data:
{cashflow_facts}

Provide caring, non-judgmental advice to Sarah:
1. Acknowledge her recent payment challenges
2. Identify payment timing patterns
3. Suggest practical solutions
4. Offer encouragement about positive behaviors
5. Provide specific next steps
"""

payment_insight = await client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": ai_prompt}]
)

Insight Pessoal Gerado:

"Sarah, percebi que você teve alguns pagamentos atrasados recentemente — totalmente compreensível depois daquele conserto inesperado de $1.200 no carro! A boa notícia é que você está com uma média de apenas 4,5 dias de atraso, o que mostra que você está se virando apesar do desafio. Vejo que sua renda normalmente cai no dia 1 e 15, mas o vencimento do seu cartão de crédito é no dia 12. Aqui está minha sugestão: configure pagamentos automáticos pelo valor mínimo no dia 2 de cada mês, quando você tem fluxo de caixa, e faça pagamentos adicionais quando sua renda freelance entrar. Além disso, você tem $2.400 em poupança de emergência — isso é muito bom! Você está indo melhor do que imagina."

Revisão Abrangente da Saúde Financeira

Passo 1: Múltiplas Buscas Direcionadas

# Search for all spending categories
spending_categories = await graphiti.search(
    "Sarah spending categories coffee groceries entertainment shopping utilities",
    limit=30
)

# Search for income and positive behaviors
positive_behaviors = await graphiti.search(
    "Sarah income 401k savings subscription cancellation meal prep positive",
    limit=20
)

# Search for stress indicators and challenges
challenges = await graphiti.search(
    "Sarah emergency expense late payment debt loan application stress",
    limit=15
)

Passo 2: Revisão Financeira Pessoal Abrangente

all_financial_data = {
    "spending": [r.fact for r in spending_categories],
    "positive": [r.fact for r in positive_behaviors],
    "challenges": [r.fact for r in challenges]
}

comprehensive_prompt = f"""
You are Sarah's personal financial AI coach. Provide an encouraging, comprehensive review of her financial situation.

Sarah's Spending Patterns:
{all_financial_data['spending']}

Sarah's Positive Behaviors:
{all_financial_data['positive']}

Sarah's Current Challenges:
{all_financial_data['challenges']}

Provide a supportive financial review for Sarah:
1. Celebrate her positive financial behaviors
2. Provide honest but encouraging assessment
3. Top 3 areas where she can improve
4. Specific action steps for next 30 days
5. Long-term financial goals alignment
6. Motivation and confidence building
"""

comprehensive_insight = await client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": comprehensive_prompt}]
)

Insight Pessoal Gerado:

"Sarah, deixa eu começar pelo que você está fazendo CERTO! Você contribui $500 mensais para o 401k (isso dá $6.000 por ano!), cancelou assinaturas proativamente para economizar dinheiro, e ainda faz meal prep — essas são atitudes de nível avançado em finanças pessoais! Seu desafio agora é que seus gastos discricionários ($1.440/mês com café + delivery de comida) estão competindo com a recuperação do seu fundo de emergência após o conserto do carro. Aqui está seu plano de ação para 30 dias: 1) Tente preparar café em casa 3 dias por semana (economize $400/mês), 2) Configure transferências automáticas de $50 semanais para reconstruir seu fundo de emergência, 3) Use sua renda freelance ($350/mês) especificamente para despesas de lazer. Você está em situação melhor do que 70% dos americanos — basta otimizar seus hábitos diários. Quer que eu crie um desafio de poupança personalizado para você?"

Alertas de Gastos e Orientação em Tempo Real

Passo 1: Busca pelos Padrões de Gastos Recentes

# Search for today's and recent spending
recent_spending = await graphiti.search(
    "Sarah spending today recent unusual large amount impulse",
    limit=10
)

# Search for budget and goals context
budget_context = await graphiti.search(
    "Sarah monthly budget goals savings target spending limits",
    limit=15
)

Passo 2: Orientação de Gastos em Tempo Real

recent_facts = [result.fact for result in recent_spending]
budget_facts = [result.fact for result in budget_context]

ai_prompt = f"""
You are Sarah's real-time financial assistant. Based on her recent spending, provide immediate, helpful guidance.

Recent Spending Activity:
{recent_facts}

Budget Context:
{budget_facts}

Provide immediate, supportive guidance:
1. Acknowledge her recent purchase
2. Put it in context of her monthly spending
3. Offer practical alternatives if concerning
4. Provide encouragement about good choices
5. Suggest next steps if needed
"""

realtime_insight = await client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": ai_prompt}]
)

Insight Pessoal Gerado:

"Ei, Sarah! Vi que você acabou de gastar $142 na Amazon em fones de ouvido e livros. Os fones são um ótimo investimento para seu setup de trabalho, e adoro que você esteja lendo livros de autoajuda! Só um aviso amigável — você está em $380 em compras online este mês, o que representa cerca de 60% do seu orçamento discricionário. Sem estresse! Talvez segurar as compras não essenciais pela próxima semana? Seu pagamento freelance deve chegar em breve, e isso pode cobrir qualquer desejo de compra restante. Continue o ótimo trabalho no desenvolvimento pessoal!"

5.4 Como o Assistente Financeiro Pessoal de Sarah Funciona

O assistente de IA de Sarah gera insights personalizados realizando primeiro buscas direcionadas no Graphiti, e depois analisando os resultados com IA para fornecer conselhos acionáveis.

Verificação Básica de Saúde Financeira Pessoal

from graphiti_core import Graphiti
from openai import OpenAI

client = OpenAI()
graphiti = Graphiti()

async def get_personal_financial_health_check(user_id: str):
    # Search for payment patterns to help Sarah stay on track
    payment_search = await graphiti.search(
        f"{user_id} payment history bills upcoming due dates",
        limit=20
    )
    
    # Search for spending patterns to identify optimization opportunities
    spending_search = await graphiti.search(
        f"{user_id} spending habits categories monthly trends",
        limit=25
    )
    
    # Search for positive behaviors to celebrate
    positive_search = await graphiti.search(
        f"{user_id} savings 401k emergency fund positive financial behaviors",
        limit=15
    )
    
    # Extract facts from search results
    payment_facts = [result.fact for result in payment_search]
    spending_facts = [result.fact for result in spending_search]
    positive_facts = [result.fact for result in positive_search]
    
    # Generate personalized financial health insight
    insight_prompt = f"""
    You are Sarah's personal financial AI assistant. Based on her financial data, provide encouraging, actionable advice.
    
    Payment History:
    {payment_facts}
    
    Spending Patterns:
    {spending_facts}
    
    Positive Behaviors:
    {positive_facts}
    
    Provide friendly, supportive advice:
    1. Celebrate her positive financial behaviors
    2. Identify her biggest opportunity for improvement
    3. Suggest one specific action for this week
    4. Provide motivation and encouragement
    5. Ask if she wants help with any specific area
    """
    
    response = await client.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": insight_prompt}]
    )
    
    return {
        "user_id": user_id,
        "insight_type": "personal_financial_health",
        "search_results_count": len(payment_search) + len(spending_search) + len(positive_search),
        "ai_advice": response.choices[0].message.content
    }

Análise Avançada por Categoria de Gastos

from graphiti_core.search.search_config_recipes import EDGE_HYBRID_SEARCH_RRF
from graphiti_core.search.search_config import SearchConfig

async def get_category_spending_advice(user_id: str, category: str):
    # Custom search configuration for edges (relationships) to find spending patterns
    search_config = EDGE_HYBRID_SEARCH_RRF.model_copy(deep=True)
    search_config.limit = 30
    search_config.mmr_lambda = 0.7  # Adjust diversity vs relevance
    
    # Search for spending relationships in this category
    spending_search = await graphiti._search(
        query=f"{user_id} spent spending {category} amount frequency habit pattern",
        config=search_config
    )
    
    # Search for income context to provide relative advice
    income_search = await graphiti.search(
        f"{user_id} income salary revenue earnings monthly",
        limit=10
    )
    
    # Search for similar category optimizations Sarah has done
    optimization_search = await graphiti.search(
        f"{user_id} saved money optimization budget reduction {category}",
        limit=10
    )
    
    # Extract relationship facts
    spending_facts = [edge.fact for edge in spending_search.edges]
    income_facts = [result.fact for result in income_search]
    optimization_facts = [result.fact for result in optimization_search]
    
    # Generate personalized spending advice
    advice_prompt = f"""
    You are Sarah's personal financial coach. Help her optimize her {category} spending.
    
    Sarah's {category} Spending Patterns:
    {spending_facts}
    
    Sarah's Income Context:
    {income_facts}
    
    Previous Optimization Efforts:
    {optimization_facts}
    
    Provide supportive, actionable advice:
    1. Acknowledge her current spending in this category
    2. Calculate percentage of income spent on {category}
    3. Identify trends (increasing/stable/decreasing)
    4. Suggest specific, realistic optimization strategies
    5. Provide motivation and celebrate any positive changes
    6. Ask if she wants help creating a plan for this category
    """
    
    response = await client.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": advice_prompt}]
    )
    
    return {
        "user_id": user_id,
        "category": category,
        "spending_patterns_found": len(spending_search.edges),
        "context_facts": len(income_search),
        "optimization_history": len(optimization_search),
        "personal_advice": response.choices[0].message.content
    }

Planejamento de Metas Financeiras com Múltiplas Buscas

async def comprehensive_financial_goals_planning(user_id: str):
    # Multiple parallel searches for different aspects of Sarah's financial life
    searches = await asyncio.gather(
        # Current financial habits
        graphiti.search(f"{user_id} spending habits daily weekly monthly routine", limit=20),
        
        # Savings and investment behaviors
        graphiti.search(f"{user_id} savings 401k investment emergency fund", limit=15),
        
        # Debt and payment patterns
        graphiti.search(f"{user_id} debt payment credit card loan balance", limit=15),
        
        # Income sources and stability
        graphiti.search(f"{user_id} income salary freelance job employment", limit=12),
        
        # Financial goals and aspirations
        graphiti.search(f"{user_id} goals plans future financial dreams", limit=10),
        
        # Positive financial behaviors
        graphiti.search(f"{user_id} budget optimization savings achievements", limit=15)
    )
    
    habits_results, savings_results, debt_results, income_results, goals_results, positive_results = searches
    
    # Organize search results by financial areas
    financial_data = {
        "spending_habits": [r.fact for r in habits_results],
        "savings_behaviors": [r.fact for r in savings_results],
        "debt_situation": [r.fact for r in debt_results],
        "income_sources": [r.fact for r in income_results],
        "financial_goals": [r.fact for r in goals_results],
        "positive_behaviors": [r.fact for r in positive_results]
    }
    
    # Generate comprehensive financial planning advice
    planning_prompt = f"""
    You are Sarah's personal financial planning assistant. Help her create a comprehensive financial plan based on her current situation.
    
    Current Spending Habits: {financial_data['spending_habits']}
    Savings & Investment Behaviors: {financial_data['savings_behaviors']}
    Debt Situation: {financial_data['debt_situation']}
    Income Sources: {financial_data['income_sources']}
    Financial Goals: {financial_data['financial_goals']}
    Positive Behaviors: {financial_data['positive_behaviors']}
    
    Provide comprehensive, encouraging financial planning advice:
    1. Celebrate her current positive financial behaviors
    2. Assess her progress toward financial stability (0-100)
    3. Identify her top 3 financial opportunities
    4. Create a prioritized action plan for next 90 days
    5. Suggest specific strategies for her biggest challenges
    6. Provide motivation and confidence-building insights
    7. Ask what specific goals she wants to focus on first
    """
    
    response = await client.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": planning_prompt}]
    )
    
    return {
        "user_id": user_id,
        "total_search_results": sum(len(results) for results in searches),
        "financial_areas": list(financial_data.keys()),
        "comprehensive_plan": response.choices[0].message.content
    }

Coaching Financeiro Pessoal em Tempo Real

async def provide_realtime_financial_coaching(user_id: str):
    # Monitor for opportunities to help Sarah with her financial decisions
    coaching_searches = {
        "spending_decision": f"{user_id} spending today recent purchase decision",
        "saving_opportunity": f"{user_id} saved money optimization achievement",
        "bill_reminder": f"{user_id} bills due upcoming payment reminder",
        "financial_goal": f"{user_id} goal progress achievement milestone"
    }
    
    for coaching_type, search_query in coaching_searches.items():
        recent_results = await graphiti.search(search_query, limit=8)
        
        if recent_results:
            # Generate immediate coaching insight
            coaching_facts = [result.fact for result in recent_results]
            
            coaching_prompt = f"""
            You are Sarah's supportive financial coach. Based on her recent activity, provide helpful guidance.
            
            Recent Activity Type: {coaching_type}
            Recent Facts:
            {coaching_facts}
            
            Provide encouraging, actionable coaching:
            1. Acknowledge her recent activity positively
            2. Provide context about her overall financial journey
            3. Suggest immediate helpful actions (if any)
            4. Offer encouragement and motivation
            5. Ask if she wants help with anything specific
            """
            
            response = await client.chat.completions.create(
                model="gpt-4",
                messages=[{"role": "user", "content": coaching_prompt}]
            )
            
            # Send personalized coaching message to Sarah
            send_coaching_message({
                "user_id": user_id,
                "coaching_type": coaching_type,
                "search_results": len(recent_results),
                "personal_coaching": response.choices[0].message.content,
                "timestamp": datetime.now().isoformat()
            })

5.5 Ingestão Simples de Eventos

import requests
import json
from datetime import datetime

def send_financial_event(user_id: str, description: str, amount: float, category: str):
    """
    Send a financial event to Graphiti for AI analysis
    """
    payload = {
        "actor": f"user:{user_id}",
        "content": description,
        "timestamp": datetime.now().isoformat(),
        "metadata": {
            "amount": amount,
            "category": category
        }
    }
    
    response = requests.post(
        "http://localhost:8000/v1/episodes",
        headers={"Content-Type": "application/json"},
        data=json.dumps(payload)
    )
    
    return response.json()

# Example usage
send_financial_event("sarah_jones", "Sarah spent $45 at Starbucks", 45.0, "coffee")
send_financial_event("sarah_jones", "Sarah missed credit card payment", 1200.0, "missed_payment")
send_financial_event("sarah_jones", "Sarah applied for personal loan", 15000.0, "loan_application")

5.6 Fluxo de Trabalho para Geração de Insights Baseada em Busca

Aqui está um fluxo de trabalho completo mostrando como usar a funcionalidade de busca do Graphiti para gerar insights:

from graphiti_core import Graphiti
from graphiti_core.search.search_config_recipes import EDGE_HYBRID_SEARCH_RRF, NODE_HYBRID_SEARCH_RRF
from openai import OpenAI
import asyncio

async def generate_personal_financial_insights(user_id: str):
    """
    Complete workflow for generating personal financial insights using Graphiti search
    """
    graphiti = Graphiti()
    client = OpenAI()
    
    # Step 1: Search for different types of financial behavior
    print(f"Searching for financial patterns for {user_id}...")
    
    # Basic search for spending patterns
    spending_results = await graphiti.search(
        f"{user_id} spending amount money paid",
        limit=25
    )
    
    # Advanced search for relationships using custom config
    relationship_config = EDGE_HYBRID_SEARCH_RRF.model_copy(deep=True)
    relationship_config.limit = 20
    
    payment_relationships = await graphiti._search(
        query=f"{user_id} payment relationship missed late",
        config=relationship_config
    )
    
    # Search for entities (nodes) related to the user
    entity_config = NODE_HYBRID_SEARCH_RRF.model_copy(deep=True)
    entity_config.limit = 15
    
    user_entities = await graphiti._search(
        query=f"{user_id} person customer account",
        config=entity_config
    )
    
    # Step 2: Extract and organize search results
    spending_facts = [result.fact for result in spending_results]
    payment_facts = [edge.fact for edge in payment_relationships.edges]
    entity_summaries = [node.summary for node in user_entities.nodes]
    
    # Step 3: Generate AI insight from search results
    insight_prompt = f"""
    Generate personalized financial coaching insights for {user_id} based on these search results:
    
    SPENDING PATTERNS:
    {spending_facts}
    
    PAYMENT RELATIONSHIPS:
    {payment_facts}
    
    USER ENTITIES:
    {entity_summaries}
    
    Please provide:
    1. Personal financial health summary
    2. Top 3 areas for improvement
    3. Top 2 positive behaviors to celebrate
    4. Specific actionable recommendations for better financial habits
    5. Motivational insights based on their progress
    """
    
    response = await client.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": insight_prompt}]
    )
    
    # Step 4: Return structured insight
    return {
        "user_id": user_id,
        "search_results": {
            "spending_patterns": len(spending_results),
            "payment_relationships": len(payment_relationships.edges),
            "user_entities": len(user_entities.nodes)
        },
        "ai_generated_insight": response.choices[0].message.content,
        "data_sources": "Graphiti search results",
        "generation_method": "search_then_analyze"
    }

# Example usage
async def run_insight_generation():
    insights = await generate_personal_financial_insights("sarah_jones")
    print(f"Generated insights based on {sum(insights['search_results'].values())} search results")
    print(insights['ai_generated_insight'])

# Run the workflow
asyncio.run(run_insight_generation())

Saída de Exemplo:

Generated insights based on 47 search results

PERSONAL FINANCIAL HEALTH SUMMARY:
Sarah shows a mix of positive and challenging financial behaviors. She demonstrates excellent retirement planning with consistent 401k contributions and income diversification through freelance work, but her daily spending habits (especially coffee purchases) represent a significant opportunity for optimization.

TOP 3 AREAS FOR IMPROVEMENT:
1. Daily coffee spending ($45/day = $990/month, 12% of income)
2. Payment timing optimization (averaging 4.5 days late, manageable but improvable)
3. Emergency fund rebuilding after recent car repair expense

TOP 2 POSITIVE BEHAVIORS TO CELEBRATE:
1. Consistent retirement savings - $500/month 401k contributions show excellent long-term planning
2. Proactive cost-cutting through subscription cancellations, saving $47/month

SPECIFIC ACTIONABLE RECOMMENDATIONS:
- Try home brewing coffee 3 days per week to save $400/month
- Set up automatic bill payments to avoid late fees
- Use freelance income specifically for emergency fund rebuilding
- Create a "fun money" budget for guilt-free discretionary spending

MOTIVATIONAL INSIGHTS:
Sarah is actually doing better than 70% of Americans with her retirement savings rate. Her main challenge is optimizing daily habits - small changes to her coffee routine could free up nearly $1,000/month for emergency savings and debt reduction. She's already shown she can cut costs (subscription cancellations) and earn extra income (freelance work).

5.7 Gerando Coaching Financeiro Pessoal com IA

from graphiti_core import Graphiti
from openai import OpenAI

client = OpenAI()
graph = Graphiti()

def fetch_personal_financial_context(user_id: str) -> str:
    """
    Retrieve Sarah's comprehensive financial context from Graphiti
    """
    return graph.get_user_context(user_id, days=90)

def generate_personal_financial_coaching(financial_context: str, user_id: str) -> str:
    """
    Generate personalized financial coaching and advice for Sarah
    """
    prompt = f"""
    You are Sarah's personal financial AI coach. Analyze her financial behavior and provide supportive, actionable guidance.
    
    User: {user_id}
    Financial Context (Last 90 Days):
    {financial_context}

    Provide encouraging, personalized coaching:
    1. Warm greeting and positive acknowledgment
    2. Top 3 financial strengths Sarah is demonstrating
    3. Her biggest opportunity for financial improvement
    4. Specific, actionable steps she can take this week
    5. Long-term financial goal suggestions based on her patterns
    6. Motivation and encouragement about her financial journey
    7. Ask what specific area she'd like help with most
    
    Format as a supportive conversation with Sarah, not a formal report.
    """
    
    response = client.chat.completions.create(
        model="gpt-4",
        messages=[
            {"role": "system", "content": "You are Sarah's supportive personal financial coach. Provide encouraging, actionable advice in a friendly tone."},
            {"role": "user", "content": prompt}
        ]
    )
    
    return response.choices[0].message.content

def generate_weekly_financial_check_in(user_id: str) -> str:
    """
    Generate a weekly financial check-in for Sarah
    """
    # Get recent financial activity
    recent_context = graph.get_user_context(user_id, days=7)
    
    prompt = f"""
    You are Sarah's weekly financial check-in assistant. Review her past week and provide supportive guidance.
    
    Sarah's Activity This Week:
    {recent_context}
    
    Provide a friendly weekly financial check-in:
    1. Celebrate any positive financial decisions from this week
    2. Identify any spending patterns worth noting
    3. Suggest one small improvement for next week
    4. Provide encouragement about her overall progress
    5. Ask about her financial goals for the coming week
    
    Keep it encouraging and actionable!
    """
    
    response = client.chat.completions.create(
        model="gpt-4",
        messages=[
            {"role": "system", "content": "You are Sarah's supportive weekly financial check-in coach."},
            {"role": "user", "content": prompt}
        ]
    )
    
    return response.choices[0].message.content

# Generate personalized coaching for Sarah
financial_context = fetch_personal_financial_context("sarah_jones")
coaching = generate_personal_financial_coaching(financial_context, "sarah_jones")
print(coaching)

# Generate weekly check-in
weekly_checkin = generate_weekly_financial_check_in("sarah_jones")
print(weekly_checkin)

Exemplo de Coaching Financeiro Pessoal Gerado pela IA:

Hi Sarah! 👋

I've been looking at your financial journey over the past 90 days, and I'm really impressed with some of the positive steps you're taking!

YOUR TOP 3 FINANCIAL STRENGTHS:
1. 🎯 Consistent 401k contributions ($500/month) - You're building a solid retirement foundation!
2. ✂️ Proactive cost-cutting - You canceled subscriptions saving $47/month. That's smart budgeting!
3. 💪 Income diversification - Your freelance work brings in extra $350/month. Great hustle!

YOUR BIGGEST OPPORTUNITY:
Your daily coffee habit ($45/day = $990/month) is 12% of your income. I know good coffee is important, but this could be your fast track to financial freedom!

THIS WEEK'S ACTION STEPS:
1. Try brewing coffee at home 3 days this week (save ~$135)
2. Set up automatic $50 weekly transfer to emergency fund
3. Track your "impulse purchases" - you had $60 unplanned at Target

LONG-TERM GOAL SUGGESTIONS:
• Build emergency fund to $5,000 (you're at $2,400 - so close!)
• Create a "fun money" budget for guilt-free spending
• Consider a side hustle goal: grow freelance to $500/month

YOU'RE DOING GREAT! Despite the car repair setback, you're managing your finances responsibly. Your payment delays are minimal (4.5 days avg), and you're actively working on solutions.

What would you like to focus on most right now? Your coffee optimization, emergency fund building, or maybe creating a monthly budget that includes fun money? I'm here to help! 💪

Exemplo de Check-in Semanal:

Hey Sarah! Time for our weekly financial check-in! 🌟

THIS WEEK'S WINS:
• You meal prepped at Costco ($85) - smart move for both health and budget!
• Invested in yourself with that Python course ($299) - career development is always a good investment!
• Only one impulse purchase this week (much better than last week!)

SPENDING PATTERNS TO NOTE:
Your coffee spending was $315 this week (7 days × $45). That's consistent with your habit, but remember our goal to try home brewing 3 days?

NEXT WEEK'S SMALL IMPROVEMENT:
Try the "24-hour rule" before any purchase over $50. This week you bought headphones ($142) - were they planned or impulse?

YOU'RE MAKING PROGRESS! 📈
Your emergency fund is slowly rebuilding, and I love seeing your focus on career development. Keep up the great work!

What's your financial goal for next week? Want to try that coffee experiment? 😊

6 Mais Exemplos de Como a IA Aprende com os Hábitos Financeiros de Sarah

Corte de Custos Inteligente

{
  "actor": "user:sarah_jones",
  "content": "Sarah canceled her Netflix subscription of $15.99",
  "timestamp": "2025-07-16T10:30:00Z",
  "metadata": {
    "action": "cancel",
    "service": "Netflix",
    "monthly_amount": 15.99,
    "reason": "cost_cutting"
  }
}

O que a IA diz para Sarah:

"Ótimo trabalho cancelando a Netflix! Você já economizou $47/mês cancelando 3 assinaturas. Isso mostra que você está levando seu orçamento a sério — continue assim! Esse corte de custos vai te ajudar a atingir sua meta de fundo de emergência mais rapidamente."

Planejamento para Aposentadoria

{
  "actor": "user:sarah_jones",
  "content": "Sarah invested $500 in index funds through her 401k",
  "timestamp": "2025-07-17T09:15:00Z",
  "metadata": {
    "amount": 500.00,
    "investment_type": "401k",
    "asset_class": "index_funds",
    "frequency": "monthly"
  }
}

O que a IA diz para Sarah:

"Você está arrasando no 401k! Mesmo com o gasto inesperado com o conserto do carro, você continua investindo $500/mês para a aposentadoria. Essa consistência demonstra uma forte disciplina financeira. Você está construindo um futuro sólido!"

Despesa de Emergência

{
  "actor": "user:sarah_jones",
  "content": "Sarah paid $1,200 for car repair at AutoFix",
  "timestamp": "2025-07-18T14:45:00Z",
  "metadata": {
    "amount": 1200.00,
    "category": "emergency",
    "merchant": "AutoFix",
    "payment_method": "credit_card",
    "urgency": "high"
  }
}

O que a IA diz para Sarah:

"Sei que aquele conserto de $1.200 no carro foi inesperado e estressante. Como você usou o cartão de crédito, vamos fazer um plano para quitá-lo rapidamente. Quer reduzir temporariamente o orçamento de café para direcionar esse dinheiro para essa despesa? Posso te ajudar a criar um plano de quitação."

8 Conclusão

O Graphiti dá à IA uma memória persistente que nunca esquece nada relevante. Pela jornada financeira de Sarah, vimos como eventos simples como "Sarah comprou café por $45" ou "Sarah pagou sua conta com 5 dias de atraso" se tornam conhecimento estruturado que o ChatGPT pode usar para dar conselhos genuinamente úteis.

O padrão é: buscar no Graphiti o contexto relevante, alimentar esse contexto ao GPT, obter conselhos específicos para essa pessoa em vez de genéricos. É isso que faz a diferença entre um chatbot que dá dicas orçamentárias genéricas e um coach que sabe que Sarah gasta $990/mês com café e que esse hábito começou seis meses atrás quando ela se mudou para o centro.

A memória está no Graphiti. O raciocínio está no GPT. Nenhum dos dois, sozinho, é suficiente.

Pronto para construir seu próprio coach financeiro com IA?
git clone https://github.com/getzep/graphiti e comece a experimentar.