Liminl Capture Service
Kushal-Kongara/liminl-capture-service
Converte mensagens e screenshots em dados estruturados com Claude
Ver no GitHubServiço de captura alimentado por IA que converte mensagens desestruturadas e screenshots do LinkedIn em ações, perfis e perguntas de onboarding estruturados. Integra Claude para extração inteligente de dados com suporte a imagens e processamento de linguagem natural.
Instalação
npx convez-ai Kushal-Kongara/liminl-capture-service1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Extrair contatos de mensagens freeform para sistemas CRM
- Processar screenshots do LinkedIn em dados de perfil estruturados
- Gerar perguntas de onboarding personalizadas baseadas em contexto
- Automatizar captura de dados em fluxos mobile
Veredito

Plugin sólido de extração de dados estruturados com Claude, bem pensado para onboarding de contatos. Código limpo, sem storage permanente, pronto para produção — ideal para quem precisa converter input desestruturado em dados tipados.
README
A tradução para o português deste README ainda está na fila.
Liminl Capture Service
A standalone AI capture service that turns messy user messages and LinkedIn profile screenshots into structured onboarding context.
Built as a trial project to explore Claude-powered data extraction for the Liminl onboarding flow.
Live link: https://liminl.vercel.app/
Features
- Message to structured action — converts freeform text ("I met Sarah from Stripe") into a typed intent, extracted params, and a graph preview
- LinkedIn screenshot to structured profile — extracts name, headline, current role, past roles, skills, education, and location from an uploaded screenshot
- Personalized follow-up questions — generates 2–3 questions grounded in specific profile facts (not generic filler)
- Graceful handling of bad inputs — blurry images, non-LinkedIn screenshots, unclear messages, and thin profiles all return clean structured error responses
- Mobile-responsive web interface — works on phone for on-the-go capture
GitHub workflow practice branch
Tech Stack
- Next.js (App Router)
- TypeScript
- Tailwind CSS
- Anthropic Claude API (
@anthropic-ai/sdk) - Vercel-ready deployment
Local Setup
npm install
Create .env.local in the project root:
ANTHROPIC_API_KEY=your_key_here
ANTHROPIC_MODEL=claude-sonnet-4-6
Start the dev server:
npm run dev
Open http://localhost:3000.
API Endpoints
GET /api/health
Returns service status.
curl http://localhost:3000/api/health
{ "ok": true, "service": "liminl-capture-service" }
POST /api/message
Converts a natural language message into a structured action.
curl -X POST http://localhost:3000/api/message \
-H "Content-Type: application/json" \
-d '{"message": "I met Sarah from Stripe"}'
{
"success": true,
"intent": "add_contact",
"params": { "name": "Sarah", "company": "Stripe", "context": "met" },
"confidence": 0.92,
"explanation": "User met a person named Sarah who works at Stripe.",
"graph_preview": {
"nodes": [
{ "type": "person", "label": "Sarah" },
{ "type": "company", "label": "Stripe" }
],
"edges": [
{ "from": "Sarah", "to": "Stripe", "type": "works_at" },
{ "from": "user", "to": "Sarah", "type": "met" }
]
}
}
POST /api/profile
Analyzes a LinkedIn profile screenshot. Accepts multipart/form-data with field image.
curl -X POST http://localhost:3000/api/profile \
-F "image=@/path/to/screenshot.png"
See docs/API_CONTRACT.md for full response shape and examples.
Data & Privacy
- No login or authentication
- No database
- No permanent storage
- Uploaded screenshots are processed in memory and immediately discarded — nothing is saved to disk or any external service
Deployment
Deploy to Vercel or any Node.js host. Set the following environment variables in your deployment environment:
ANTHROPIC_API_KEY=your_key_here
ANTHROPIC_MODEL=claude-sonnet-4-6
ANTHROPIC_MODEL is optional and defaults to claude-sonnet-4-6 if not set.
Project Docs
| Doc | Description |
|---|---|
| docs/PROJECT_SCOPE.md | What this project does and does not do |
| docs/API_CONTRACT.md | Full endpoint contracts and response shapes |
| docs/PROMPTS.md | Claude prompt strategy and templates |
| docs/TEST_CASES.md | Manual test cases |