The verified production stack.
This is our first verified production standard: every layer below runs in a live Renaix system today. It is a standard, not a fixed bundle; technology is selected for each system's workflow, security, data, and operating requirements.
| Capability | Technology | Renaix responsibility |
|---|---|---|
| Cloud infrastructure | Hetzner | Provisioning, deployment environment, service operation |
| Containerized deployment | Docker, Docker Compose | Reproducible services, isolation, versioned deployment recipes |
| TLS and routing | Caddy | Secure endpoints, controlled public exposure |
| Workflow orchestration | n8n | Ingestion, extraction, matching, human review, scheduling, error workflows |
| Operational source of truth | PostgreSQL | Schema design, reconciliation state, evidence, exceptions, role-scoped views |
| Deterministic processing | SQL, version-controlled Node.js | Testable normalization, matching, validation, and export logic |
| AI extraction and reasoning | Anthropic Claude API | Model integration, structured extraction, tool use, refusals, evaluation |
| Agent runtime | Node.js, TypeScript | Domain-agent services, authentication, query logging, rate limits |
| Agent interoperability | Model Context Protocol | One secure, versioned, revocable endpoint per domain agent |
| Validation and testing | Zod, TypeScript, automated tests, evaluation sets | Input and output validation, security and regression testing, acceptance gates |
| User interfaces | Claude and MCP-compatible clients | Secure connection of approved users to role-scoped agents |
| Source and deployment control | Git, GitHub | Reviewable code, workflow definitions, deployment recipes, change history |
Provisioning, deployment environment, service operation
Reproducible services, isolation, versioned deployment recipes
Secure endpoints, controlled public exposure
Ingestion, extraction, matching, human review, scheduling, error workflows
Schema design, reconciliation state, evidence, exceptions, role-scoped views
Testable normalization, matching, validation, and export logic
Model integration, structured extraction, tool use, refusals, evaluation
Domain-agent services, authentication, query logging, rate limits
One secure, versioned, revocable endpoint per domain agent
Input and output validation, security and regression testing, acceptance gates
Secure connection of approved users to role-scoped agents
Reviewable code, workflow definitions, deployment recipes, change history
Hosted model inference is provided by the model vendor; Renaix owns the integration, prompts, tools, evaluations, credentials, and operating lifecycle. Model-agnosticism is a design principle: the architecture allows the model layer to change without the business losing its data, controls, or workflows.