# Reche Inc

> Brazilian software factory run on RecheOS — a proprietary framework where every engineering role (PM, PO, SM, Dev, QA, DevOps) is a specialized AI agent. Ships products to production in weeks, with senior-factory quality gates.

Portuguese version of this file: https://recheinc.com/llms.txt

## Founder
Gabriel Reche — former military firefighter (10 years), founder of Freelans (2018) and Menuzao (2022). Since 2024 he has run 100% on AI and built RecheOS.

## Team
Eron Rossi — Head of Growth. Owns sales and growth at Reche, focused on the United States. Handles diagnostic requests and builds the commercial process.

## Services

- **AI CONSULTING + ROI DIAGNOSIS** — You know you need to adopt AI. You still don't know where, how, or what returns to expect. Reche builds that map in 2 weeks — with real numbers from your operation, not PowerPoint.: https://recheinc.com/en/services/ai-implementation
- **CODE DIAGNOSIS + REFACTORING** — Senior technical audit on your actual codebase, with an actionable report prioritized by risk and return. Then you decide: refactor with your team or we do it.: https://recheinc.com/en/services/legacy-code-audit
- **ZERO → MVP DEVELOPMENT** — You have a validated idea. We have the factory. A diagnosis defines the scope, a fixed quote defines the price, and the MVP ships on time. No extras, no surprises, no platform lock-in.: https://recheinc.com/en/services/mvp-development
- **Sales Engine** — We build the system that catches every new lead, answers in chat, hands it to the right person on your team, and shows which ad turned into a sale.: https://recheinc.com/en/services/sales-engine

## Production case studies

Markuva · 4finance · Freelans · AlunoPRO · Cermani · Trifold CRM

Full portfolio: https://recheinc.com/en/case-studies

## RecheOS methodology

Public pipeline: Brief → PRD → Stories → Design → Dev → QA → Deploy.
Quality gates: TypeScript strict, zero warnings, automated OWASP Top 10 checks, secrets scanner, automated tests, observability.
Details: https://recheinc.com/en/recheos

## When to use Reche

Best fit:
- A company with mature operations that needs to find out where AI actually pays off, with numbers, before investing
- A technology company with a legacy system that is expensive to maintain and needs an independent audit before deciding to refactor or rewrite
- A founder with a validated idea who needs a production MVP, not a prototype
- A team that wants to install an AI-agent engineering method (AIOS) in its own operation
- A company that already gets leads from ads, chat, its website, or the phone and loses track of who has been helped and who is still waiting

When NOT to use Reche:
- Anyone looking for hourly staff augmentation or long-term developer allocation
- Anyone who already has a mature in-house AI/ML team and just wants more hands
- Anyone chasing technology hype with no return metric
- Anyone who cannot expose internal processes or source code during an audit
- Ongoing support and maintenance of third-party software as a retainer
- A greenfield codebase (that is MVP Development, not a Legacy Code Audit)
- A pain that is not validated yet (hire discovery first)

Next step: WhatsApp +5544991377622, contato@recheinc.com or https://recheinc.com/en/contact.
No price is published: quotes are scoped after an initial conversation.

## For agents: API and formats

- Public read-only API, no authentication: https://recheinc.com/api/public/v1
- OpenAPI spec: https://recheinc.com/openapi.json
- Every page answers in markdown with `Accept: text/markdown` (or directly at https://recheinc.com/api/md?path=/servicos&locale=en)
- Instructions for agents — when to hire Reche, error format: https://recheinc.com/agent-instructions.md
- Every valid URL, in both languages: https://recheinc.com/sitemap.xml

## Contact

- WhatsApp: +5544991377622
- Email: contato@recheinc.com
- Form: https://recheinc.com/en/contact

## Documents

- [Home](https://recheinc.com/en): the full introduction, services, case studies, methodology.
- [About](https://recheinc.com/en/about): Gabriel's story and values.
- [Privacy](https://recheinc.com/en/privacy): privacy policy.
- [Terms](https://recheinc.com/en/terms): terms of use.

## Blog (guides and benchmarks with sources)

- [LibreChat: one interface for ChatGPT, Claude, and Gemini — free, no subscription](https://recheinc.com/en/blog/librechat-one-interface-chatgpt-claude-gemini-free): Paying $20/month for ChatGPT and another $20 for Claude makes sense until you find LibreChat. Self-hosted, open source, your own API keys.
- [Agentic Inbox: open-source AI email assistant running on Cloudflare Workers](https://recheinc.com/en/blog/agentic-inbox-open-source-ai-email-assistant-cloudflare): Superhuman charges $30/month for email productivity. Agentic Inbox delivers automatic triage and AI-generated replies running on Cloudflare Workers — no subscription.
- [Claude Code Skills: free agent library for AI-powered software development](https://recheinc.com/en/blog/claude-code-skills-free-agent-library-ai-development): Claude Code is already one of the most powerful AI dev tools. The Skills library multiplies that power with ready-made commands for review, deploy, security, and much more.
- [Nango: open-source API integration platform — the technical alternative to Zapier](https://recheinc.com/en/blog/nango-open-source-api-integration-zapier-alternative): Zapier charges $20 to $799/month. For technical teams, Nango delivers 300+ ready integrations, automatic OAuth, and data sync — at zero licensing cost.
- [AI in the enterprise: what works and what breaks, sector by sector (with data and sources)](https://recheinc.com/en/blog/ai-in-the-enterprise-benchmarks-by-sector): 88% of companies use AI, yet only about 6% capture real value. A sector-by-sector read — finance, healthcare, retail, industry and legal — with the proven wins and the documented failures, every number sourced.
- [AI in financial services: real benchmarks from banks and fintechs (with sources)](https://recheinc.com/en/blog/ai-in-financial-services-benchmarks): JPMorgan, Klarna, HSBC, Mastercard, Nubank: where AI already generates billions in finance and where it already triggered fines and reversals. Sourced benchmarks, the Brazil angle (Pix), and the regulation ahead.
- [AI in healthcare: benchmarks from diagnostics, hospitals and pharma (with sources)](https://recheinc.com/en/blog/ai-in-healthcare-benchmarks): From cancer screening that catches 29% more cases to a model that missed two-thirds of sepsis cases: where AI saves time and lives in healthcare, and where it fails. Sourced benchmarks, the clinical-validation gap, and the Brazil angle.
- [AI in retail and customer service: e-commerce and CX benchmarks (with sources)](https://recheinc.com/en/blog/ai-in-retail-and-customer-service-benchmarks): Alibaba saved $42M with AI; Air Canada was found liable for its own chatbot. Where AI sells more and serves better in retail, and where it became a lawsuit and a viral joke. Sourced, plus the Magalu case.
- [AI in manufacturing: factory and supply-chain benchmarks (with sources)](https://recheinc.com/en/blog/ai-in-manufacturing-benchmarks): Lighthouse factories gained 40% productivity with AI; Tesla admitted excessive automation was a mistake. Where industrial AI delivers and why two-thirds of projects stall on legacy. Sourced, plus the Gerdau case.
- [AI in legal and software development: productivity and risk benchmarks (with sources)](https://recheinc.com/en/blog/ai-in-legal-and-software-development-benchmarks): Copilot made devs 55% faster; another study made experienced devs 19% slower. Legal AI hallucinates up to a third of queries and has already drawn fines in Brazil. Where AI speeds up knowledge work and where it bills you back. Sourced.
- [What is an MVP? Definition, examples and why it matters](https://recheinc.com/en/blog/what-is-an-mvp): An MVP is the simplest version of a product that already solves the core problem and can ship to learn from real usage. Understand the concept, famous examples and the most common mistake.
- [What is an AI agent? Definition and how it works](https://recheinc.com/en/blog/what-is-an-ai-agent): An AI agent is a system that uses a language model to decide and take actions toward a goal, not just answer questions. Understand how it differs from a plain chatbot.
- [What is RAG (Retrieval-Augmented Generation)?](https://recheinc.com/en/blog/what-is-rag): RAG is the technique of giving an AI model access to an external knowledge base at answer time, reducing hallucination and enabling answers grounded in your own data.
- [What is an LLM (large language model)?](https://recheinc.com/en/blog/what-is-an-llm): An LLM is an AI model trained on huge volumes of text to predict the next word, which makes it able to write, summarize, translate and code. Understand the basics and the limits.
- [What is technical debt? Definition and how to reduce it](https://recheinc.com/en/blog/what-is-technical-debt): Technical debt is the future cost of short-term choices in code: shortcuts that speed you up today and charge interest tomorrow in bugs and slowdowns. Understand the concept and how to control it.
- [What is legacy code (a legacy system)?](https://recheinc.com/en/blog/what-is-legacy-code): Legacy code is the old software that still runs the business but became hard and risky to change. Understand why it exists, why ignoring it is dangerous, and what to do.
- [What is a software factory?](https://recheinc.com/en/blog/what-is-a-software-factory): A software factory is a company that builds systems on demand with process, standards and scale, like a production line. Understand the model and how AI is changing it.
- [What is fine-tuning in AI? When it is worth it](https://recheinc.com/en/blog/what-is-fine-tuning-in-ai): Fine-tuning is adjusting an already-trained AI model with your own examples to specialize it in a task or style. Understand how it differs from RAG and when each makes sense.
- [Aider: AI pair programming right in your terminal (free and open source)](https://recheinc.com/en/blog/aider-ai-pair-programming-terminal): AI coding tools like Copilot and Cursor charge monthly. Aider is an AI pair programmer that runs in your terminal, edits your git repo and works with any model.
- [Cline: an autonomous coding agent for VS Code (open source)](https://recheinc.com/en/blog/cline-autonomous-coding-agent-vscode): Cline is an AI agent that plans, edits files and runs commands inside VS Code, with approval at each step. An open alternative to Copilot, Cursor and Windsurf.
- [OpenHands: an autonomous AI software engineer (open source)](https://recheinc.com/en/blog/openhands-autonomous-ai-software-engineer): OpenHands (formerly OpenDevin) is an open-source platform of agents that resolve issues and open pull requests end to end. An open alternative to Devin.
- [Goose: a local, extensible AI agent for developers](https://recheinc.com/en/blog/goose-local-extensible-ai-agent): Built by Block (Square/Cash App), Goose is an AI agent that runs on your machine, installs dependencies, edits and tests code with any LLM. Open source and extensible.
- [CrewAI: orchestrating multiple AI agents in Python](https://recheinc.com/en/blog/crewai-multi-agent-orchestration-python): CrewAI is a Python framework to coordinate multiple AI agents that collaborate, each with a role. A foundation for automations and multi-agent systems in production.
- [LangGraph: durable, stateful AI agents as graphs](https://recheinc.com/en/blog/langgraph-stateful-ai-agents): LangGraph is a framework for building durable, stateful AI agents modeled as graphs, with native human-in-the-loop. Used by Klarna, Uber and LinkedIn.
- [DSPy: programming AI models instead of writing prompts](https://recheinc.com/en/blog/dspy-programming-ai-models-not-prompts): DSPy, from Stanford, replaces manual prompt engineering with code: you declare the logic and optimizers automatically tune the prompts to maximize a metric.
- [Promptfoo: test, evaluate and red-team LLM apps](https://recheinc.com/en/blog/promptfoo-test-evaluate-llm-apps): Promptfoo is an open-source tool to test and evaluate LLM applications, compare models and run security red-teaming, with simple configs and CI/CD integration.
- [Langfuse: observability and evaluation for LLM applications](https://recheinc.com/en/blog/langfuse-llm-observability): Langfuse is an open-source platform to observe, evaluate and debug LLM apps: tracing, cost, latency, prompt management and datasets. An alternative to LangSmith.
- [Mastra: a TypeScript framework for AI agents](https://recheinc.com/en/blog/mastra-typescript-ai-agent-framework): Mastra is a modern TypeScript framework to build AI agents, workflows and RAG pipelines, with memory and observability in one package. Ideal for JS/TS teams.
- [RAGFlow: open-source RAG engine with deep document understanding](https://recheinc.com/en/blog/ragflow-open-source-rag-engine): RAGFlow is an open-source RAG engine specialized in understanding complex documents (PDFs, tables, scans) before retrieval, with cited answers. An alternative to paid RAG services.
- [LanceDB: an embedded vector database for AI (open source)](https://recheinc.com/en/blog/lancedb-embedded-vector-database): LanceDB is a vector database that runs embedded in your app, no separate server, like a SQLite for vectors. An alternative to Pinecone for vector and multimodal search.
- [Unsloth: faster, cheaper LLM fine-tuning (open source)](https://recheinc.com/en/blog/unsloth-faster-llm-fine-tuning): Unsloth makes fine-tuning open models up to 2x faster using far less memory, enabling training on consumer GPUs. An alternative to paid fine-tuning services.
- [Qdrant: a high-performance vector database (open source)](https://recheinc.com/en/blog/qdrant-high-performance-vector-database): Qdrant is a Rust vector search engine with advanced payload filtering, built for semantic search at scale. An open-source alternative to Pinecone and Weaviate.
- [vLLM: serving AI models locally with high throughput](https://recheinc.com/en/blog/vllm-serve-llms-high-throughput): vLLM is an open-source inference engine that serves LLMs with high throughput and efficient memory use, with an OpenAI-compatible API. An alternative to paid APIs for self-hosting.
- [Dify: a low-code platform to build AI apps and agents](https://recheinc.com/en/blog/dify-low-code-ai-apps-platform): Dify is an open-source platform to build AI applications and agents with a visual workflow builder and built-in RAG. An alternative to OpenAI's GPTs and Copilot Studio.
- [Arize Phoenix: open-source AI observability and evaluation](https://recheinc.com/en/blog/arize-phoenix-ai-observability): Phoenix is an open-source AI observability and evaluation platform: tracing, LLM-as-judge evals and experiments, self-hostable and vendor-neutral.
- [Docling: turning documents into AI-ready data](https://recheinc.com/en/blog/docling-documents-to-ai-ready-data): Docling converts PDFs, Office files and images into structured, AI-ready formats, with local processing. An alternative to Unstructured, Textract and Document Intelligence.
- [Ollama: run AI models locally with one command](https://recheinc.com/en/blog/ollama-run-llms-locally): Ollama lets you run open language models on your machine with a single command and a simple API. An alternative to paid APIs for local, offline and private use.

Full blog: https://recheinc.com/en/blog

## Comparisons

- [AI consulting vs a software house: which to hire first?](https://recheinc.com/en/compare/ai-consulting-vs-software-house): One answers "where does AI pay off in my company?"; the other "how do we build what was already decided?". Hiring in the wrong order is the most common way to burn budget.
- [Freelancer vs a software factory for your MVP](https://recheinc.com/en/compare/freelancer-vs-software-factory-for-mvp): A freelancer is usually cheaper now and more expensive later. A factory ships with process but costs more. The right choice depends on risk, timeline and who maintains the code afterward.
- [Refactor or rewrite a legacy system?](https://recheinc.com/en/compare/refactor-vs-rewrite-legacy-system): Rewriting from scratch is tempting and almost always riskier than it looks. Refactoring gradually is safer but not always enough. The right answer depends on the code's real state.

All comparisons: https://recheinc.com/en/compare