iSolUso prático: o dia da operação resumido no WhatsApp, às seis da tarde
Chamados, faturamento, entregas e pessoas em seis linhas. E a política interna respondida a partir do material oficial.
ENGINEERING
Notes from the infrastructure
Uso prático: o traceback de produção vira causa, correção e teste
Cola o erro no terminal e recebe hipótese, patch e o teste que prova, sem o código sair da sua infraestrutura.
Uso prático: a função de parcelamento que perdia centavos
A revisão encontra o erro de arredondamento, explica o impacto em dinheiro e devolve a correção.
Uso prático: o assistente do nosso próprio site respondendo a objeção mais difícil
O chat que está nesta página responde sobre soberania e transforma pergunta de preço em conversa com gente.
Uso prático: fechar o mês de despesas dentro da planilha
A equipe de finanças classifica o extrato inteiro com uma fórmula. Cinco linhas em seis segundos, sem sair do Sheets.
Uso prático: transformar comentário de pesquisa em coluna que o dashboard filtra
Cada resposta aberta do NPS vira tema, sentimento e ação sugerida durante o refresh.
Uso prático: triar o e-mail de suporte antes de qualquer humano abrir
Fila, prioridade, resumo e prazo em dois segundos, direto na caixa compartilhada.
Uso prático: da nota fiscal no e-mail para os campos do contas a pagar
O anexo chega e vira fornecedor, número, datas e valor em JSON, sem digitação.
Uso prático: a thread de trinta mensagens vira seis linhas de decisão
Decisões, pendências com responsável e o risco em aberto, pedidos no próprio canal.
Uso prático: o lead que escreve às 23h não espera até amanhã
O assistente responde, qualifica com uma pergunta e informa quando o humano assume.
Uso prático: a senha que entrou no código não passa do push
Um revisor automático marca o achado de segurança antes de o código circular.
Uso prático: o alerta do plantão chega com diagnóstico, não só com número
Causa provável, primeiro comando de verificação e a decisão de acordar ou não alguém.
Hermes Agent: your operations, answered on WhatsApp
What Hermes answers, how to ask well, and what stays inside your perimeter. A usage guide for teams that already have it running.
VS Code on your own gateway: the editor everyone already has
One JSON block in the Continue extension points VS Code at your sovereign endpoint. No new editor, no new license, same shortcuts.
Zapier with your own gateway: AI in any Zap, sovereign
One Webhooks action turns any Zap into an AI step that answers from infrastructure you control.
Google Sheets: AI as a spreadsheet formula, on your own gateway
Ten lines of Apps Script turn your sovereign endpoint into =ISOL(), so the business team classifies data without writing code.
Case study: 546 commands in a closed desktop ERP, zero failures
How five agents and an orchestrator reconcile inside QuickBooks Desktop, a system with no REST API, and how trust went from 75% to 90%.
Case study: 24/7 urgent-ticket triage that calls a human until one answers
Agents watch the inbox of an industrial services company, judge urgency and escalate by SMS, WhatsApp and phone. In production, integrated with the field system.
Case study: a pre-sales agent that quotes with dollar precision
A construction company's agent talks to real leads, extracts a structured quote request and matches the official spreadsheet to the dollar.
Case study: the campaign art on this site never left the building
Our own marketing runs on iSol Media: product shots, architecture scenes and the demo video were generated on sovereign infrastructure, from the chat.
A WhatsApp assistant answered by your own gateway
Meta Cloud API webhook + one function that calls your sovereign endpoint. The handler below answered a real message in Portuguese.
A Telegram bot answered by your own gateway
Forty lines of Python, long-polling, no public URL needed. The handler below ran for real against a sovereign gateway.
AI code review in GitHub Actions, answered by your own gateway
A 30-line workflow that reviews every push with a model on sovereign infrastructure. Public repo, real runs, nothing mocked.
One gateway, every AI tool your team already uses
Terminal agent, editor, automation, BI, chat, each one opens its own channel to a vendor. Put a single OpenAI-compatible gateway in front of all of them instead.
Run Aider against your own OpenAI-compatible gateway
Two environment variables point Aider's pair-programming sessions at infrastructure you control. Your code never leaves through a channel you can't inspect.
Run Claude Code on your own OpenAI-compatible gateway
Point Claude Code at your own gateway with two environment variables. Keep your code and prompts inside infrastructure you control.
Point Cursor at your own OpenAI-compatible gateway
Two settings in Cursor route chat and Composer through your own gateway. What that covers, what it doesn't, and how to verify it.
n8n + your own gateway: AI in every workflow, one endpoint
One Base URL field in n8n's OpenAI credential routes every AI node through your own gateway. Why that matters more in automation than in chat.
OpenCode with your own gateway: config-first, no lock-in
OpenCode was built for custom providers. Point opencode.json at your own gateway and every layer of the stack stays something you could run yourself.
Add AI to Power Automate without AI Builder credits
Wire Power Automate's HTTP action to your own OpenAI-compatible gateway and skip AI Builder's prepaid credit packs entirely.
LLM enrichment in Power BI refreshes, without Fabric capacity
Call your own OpenAI-compatible gateway from Power Query with Web.Contents and enrich columns at refresh time, no Fabric F64 capacity required.
A Teams assistant on your own gateway, no Copilot seat required
Build a channel assistant in Microsoft Teams with Power Automate's HTTP action and your own gateway, no Copilot per-seat license.
A website chatbot on your own gateway, every visitor question stays home
Point Typebot (or Dify, Flowise, AnythingLLM) at your own OpenAI-compatible gateway and embed the widget on your site. Real steps, smoke test at the end.