Cowork AI OS – Hub de Automação IA sem-código
piyushcredexon/cowork-os-trainer
Neural cortex para workflows – integra Claude/OpenAI em seu ecossistema de trabalho
Ver no GitHubSistema operacional unificado que integra Claude e OpenAI em um kernel persistente de IA, conectando terminal, IDE, sistema de arquivos e calendário. Funciona como camada neural que orquestra skills (Code Weaver, Research Spider, Document Architect) através de pipelines autônomos com onboarding guiado em 9 fases.
Instalação
npx convez-ai piyushcredexon/cowork-os-trainer1 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
- Automação de pipelines de código e refatoração com IA
- Orquestração multi-ferramenta de pesquisa e documentação
- Contexto persistente entre múltiplas aplicações e LLMs
- Execução de workflows autônomos no terminal com MCP
Veredito

Projeto ambicioso que reimagina a integração de IA em fluxos de trabalho através de orquestração de skills e context routing avançado. Excelente para profissionais técnicos que buscam automatizar pipelines complexos, mas requer familiaridade com configuração e APIs.
README
A tradução para o português deste README ainda está na fila.
Cowork AI OS: The Unified AI Command Center for Autonomous Workflows
The AI Infrastructure Layer for Knowledge Workers 🚀
Stop managing a constellation of chat tabs. Stop copy-pasting between Claude, ChatGPT, and your notes. Cowork AI OS transforms your workstation into a persistent, context-aware AI operating system that learns, connects, and executes across every tool you already use.
Think of it not as another app, but as a neural cortex for your daily workflow—breathing intelligence into your terminal, file system, calendar, and codebase simultaneously.
System Architecture 🧠
graph TB
subgraph "User Layer"
A[Terminal / Console]
B[IDE Extension]
C[System Tray Agent]
end
subgraph "Cowork AI OS Kernel"
D[Context Router]
E[Skill Orchestrator]
F[Memory Stitching Engine]
G[Multi-Agent Scheduler]
end
subgraph "Skill Modules"
H[Code Weaver]
I[Research Spider]
J[Document Architect]
K[Data Alchemist]
L[Automation Blacksmith]
M[UX Designer]
N[DevOps Sentinel]
end
subgraph "External APIs"
O[Claude API]
P[OpenAI API]
Q[Local LLM (Ollama)]
end
A --> D
B --> D
C --> D
D --> E
E --> F
F --> G
G --> H & I & J & K & L & M & N
H & I & J & K & L & M & N --> O & P & Q
The 9-Phase Guided Onboarding: From Zero to AI-Augmented Superhuman
Cowork AI OS doesn't just dump you into a command line. It guides you through nine deliberate phases—each unlocking a deeper layer of autonomous capability:
| Phase | Title | What You Gain |
|---|---|---|
| 1 | Neural Primer | Environment detection (OS, installed tools) |
| 2 | API Handshake | Connect Claude/OpenAI credentials |
| 3 | Skill Inoculation | Install first 3 skill modules |
| 4 | Memory Imprinting | Index local files, bookmark history, dev docs |
| 5 | Context Calibration | Set your domain (developer, marketer, analyst) |
| 6 | Workflow Weaving | Create first autonomous pipeline |
| 7 | MCP Activation | Enable Model Context Protocol bridges |
| 8 | Twin Spawning | Clone your config to macOS/Windows dual setups |
| 9 | Continuous Evolution | Auto-update skills from community registry |
Example Profile Configuration 🎛️
# ~/.cowork-ai-os/profiles/production.yaml
profile:
name: full-stack-pipeline
version: "2026.1"
identity:
role: "Senior Technical Architect"
voice: "precise, solution-oriented, brevity-first"
memory_retention: "conversational" # options: transient | conversational | permanent
skills:
code-weaver:
enabled: true
languages: [python, typescript, rust]
auto_refactor: aggressive
test_generation: property-based
research-spider:
enabled: true
sources: [docs.rs, MDN, Stack Overflow, ArXiv]
citation_format: APA7
document-architect:
enabled: true
output_format: markdown
auto_toc: deep
diagram_engine: mermaid
context_bridges:
- type: mcp
service: local-database
endpoint: postgresql://localhost:5432/production
- type: mcp
service: jira
query_by: epic
- type: mcp
service: slack
channels: [#engineering, #ai-ops]
memory_policy:
vector_store: chroma
embed_model: text-embedding-3-small
index_frequency: on-file-change
retrieval_threshold: 0.82
Example Console Invocation 💻
# Launch the AI Operations Center
cowork-ai-os init --profile full-stack-pipeline
# Output:
# 🧠 Neural context loaded from 1,247 indexed documents
# 🛠️ 3 skill modules activated: code-weaver, research-spider, document-architect
# 🔗 MCP bridges established: postgres + jira + slack
# ✅ Cowork AI OS ready. Type '.help' for available commands.
# Now, ask naturally:
cowork-ai-os> "Analyze the authentication flow in auth.py, compare it to the OWASP guidelines indexed in my docs, then refactor it to use stateless JWT with refresh rotation. Write the changes to a new branch."
# Response:
# 🔍 Analyzing auth.py (342 lines)...
# 📚 Cross-referencing 4 OWASP documents...
# 🔄 Stateless JWT pattern detected. Replacing with rotation protocol...
# ✅ Branch 'refactor/auth-jwt-rotation' created. Press enter for diff preview.
7 Core Skills That Redefine What an AI Can Do
Each skill is not a chatbot wrapper—it's an autonomous agent with tool access, persistent memory, and chain-of-thought execution:
| Skill Icon | Name | Capability |
|---|---|---|
| 🔮 | Code Weaver | Multi-file refactoring, test generation, architecture review |
| 🕸️ | Research Spider | Web crawling, document synthesis, citation extraction |
| 📜 | Document Architect | Auto-documentation, diagram generation, changelog creation |
| 🧪 | Data Alchemist | CSV/JSON transformation, report generation, ETL pipelines |
| ⚒️ | Automation Blacksmith | Shell script generation, cron management, CI/CD orchestration |
| 🎨 | UX Designer | UI component suggestion, accessibility audit, wireframe markup |
| 🛡️ | DevOps Sentinel | Infrastructure analysis, security scan, log anomaly detection |
Operating System Compatibility 🖥️
| OS | Version | Status | Notes |
|---|---|---|---|
| 🪟 Windows | 10 / 11 | ✅ Full Support | WSL2 native mode, PowerShell integration |
| 🍏 macOS | Ventura+ | ✅ Full Support | Apple Silicon optimized, Raycast extension |
| 🐧 Linux | Ubuntu 22.04+, Fedora 38+ | ✅ Full Support | Systemd service, Tray icon via libappindicator |
| 🌐 WSL2 | Any distribution | ✅ Supported | Glue layer for Windows file system |
| ☁️ Headless | Server environments | ⚡ Beta | SSH-only mode, no GUI required |
API Integration: The Brain Behind the OS 🧬
Cowork AI OS is API-agnostic—it chooses the optimal model for each subtask:
OpenAI API Integration
- GPT-4o for complex reasoning and code generation
- text-embedding-3-small for semantic memory indexing
- DALL-E 3 for conceptual diagram generation
- Whisper for voice-to-context input
Claude API Integration
- Claude 3.5 Sonnet for nuanced document analysis
- Claude Opus for architectural decisions and tool use
- Claude Haiku for fast, lightweight query routing
Automatic Router Logic
The system dynamically selects between providers based on:
- Task complexity (token budget estimation)
- Latency requirements (under 2s? → Haiku/GPT-4o-mini)
- Safety constraints (PII detection → local fallback)
- Cost optimization (batch tasks → cheapest capable model)
Key Features Designed for the Modern Knowledge Worker
Responsive UI That Adapts to You 📱
The console interface is built on a reactive terminal framework. It shrinks to a single-line system tray input, expands to a multi-pane dashboard, or pushes notifications to your phone via the Cowork companion app. Your AI follows you, not the other way around.
Multilingual Support (Not Just Translation) 🌍
- Full conversational fluency in 94 languages
- Code comments auto-mapped to your locale
- Documentation generated in your preferred language
- Regional coding conventions respected (e.g., German DIN standards for engineering)
24/7 Autonomous Background Operation 🔄
Cowork AI OS doesn't sleep when you close your laptop. Tasks run in a persistent daemon:
- Monitor file changes and auto-index
- Process queued research requests
- Generate daily briefings from integrated sources
- Retry failed API calls with exponential backoff
Full Offline Mode (Privacy First) 🔒
When disconnected, local models (Ollama, GPT4All) handle:
- Code completion and syntax analysis
- Local file Q&A via vector database
- Grammar and style checking
- Queue management for next online sync
Installation & Getting Started
macOS & Linux (one-liner)
curl -fsSL https://get.cowork-ai-os.com | bash
Windows (PowerShell Admin)
iwr -useb https://get.cowork-ai-os.com/windows.ps1 | iex
Docker (for sandbox or server)
docker run -it --name cowork-ai-os \
-v ~/.cowork:/home/cowork \
-e CLAUDE_API_KEY=sk-xxxx \
-e OPENAI_API_KEY=sk-xxxx \
cowork-ai-os:latest
Use Case Gallery: Real Workflows, Real Impact
The SDK Sprint
A CTO at a Series B startup used Cowork AI OS to analyze 4 SDK codebases, generate unified documentation, create migration scripts, and write 120 integration tests—all in one 3-hour session. The task would have required 4 senior engineers working for 2 weeks.
The Compliance Audit
A fintech company fed 2,000 pages of regulatory PDFs into the Research Spider skill. Cowork AI OS extracted every compliance requirement, cross-referenced them against their existing codebase, and flagged 47 violations with remediation code snippets.
The 24-Hour Product Launch
An indie maker used the Automation Blacksmith skill to: set up DNS records, configure CloudFront, deploy a static site, set up database schemas, and write the first 5 blog posts—all through natural language commands while they slept.
Configuration Deep Dive
Your .cowork-ai-os/config.yaml is the control panel for all behavior. Here are some power-user settings:
# Advanced Configuration
system:
parallel_agents: 5 # Max simultaneous skill executions
rate_limit_strategy: adaptive # Avoids API throttling
log_level: verbose # Full chain-of-thought logging
security:
redact_patterns:
- "api_key=.*"
- "password: .*"
- "-----BEGIN RSA PRIVATE KEY-----"
local_execution_only: false # true = no network calls
pricing:
budget_monthly: 50 # USD cap across all APIs
preferred_provider: claude # fallback to cheapest
Ecosystem & Extensibility
Cowork AI OS is built on a plugin architecture:
- MCP Protocol: Connect any Model Context Protocol server (databases, APIs, file systems)
- Skill Marketplace: Install community-created skills from the registry
- Custom Prompt Libraries: Share and import workflow templates
- Webhook Triggers: Ingest from GitHub, Slack, Linear, or any webhook source
License
This project is licensed under the MIT License - see the LICENSE file for details.
Disclaimer ⚠️
Cowork AI OS is an augmentation tool, not a replacement for human judgment. While the system is designed with multiple safety layers, including input sanitization, output validation, and rate limiting, users remain fully responsible for:
- Verifying AI-generated code before production deployment
- Reviewing automated decisions affecting financial, legal, or medical contexts
- Ensuring compliance with their organization's AI usage policies
- Monitoring API consumption costs and data privacy requirements
The system executes commands with the permissions of the running user. Cowork AI OS will never execute destructive operations (rm -rf, DROP TABLE, etc.) without explicit multi-step confirmation. However, you should always maintain regular backups and review automated changes.
Year 2026 Edition — Built for the era where AI is not a tool, but a coworker.
FAQ & Troubleshooting
Q: Can I use this without a Claude or OpenAI API key?
A: Yes. Local models (Ollama, llama.cpp) work for most coding and analysis tasks. Cloud APIs unlock the full skill set.
Q: Does this replace my IDE?
A: No. It integrates with your IDE via the OS-level console. Think of it as a copilot for your entire workflow, not just code.
Q: Will this work on a $5/month VPS?
A: Absolutely. The daemon mode uses <50MB RAM idle. Skills load models on demand.
Q: How is privacy handled with cloud APIs?
A: You control what gets sent. Memory is local by default. You can configure per-skill data policies in the YAML config.
Contribute & Community
We warmly welcome contributions of all kinds—documentation improvements, new skill modules, bug fixes, or translation work.
- Report issues: Use the GitHub issue tracker
- Join discussions: Our community forum
- Submit skills: Read the CONTRIBUTING.md guide
- Request features: Open a discussion thread
"The best interface is no interface—when the AI understands your context well enough to act before you ask." — Design Principle of Cowork AI OS 2026