Platform architecture
Orbis Studio runs entirely within your organization's infrastructure — on Kubernetes, OpenShift, or any container-based environment. No data leaves your network unless you explicitly connect an external service.
Contact the Orbis team at orbislabs.ai. Platform installation completes in 1–3 business days once the target infrastructure is ready.
Architecture overview
A production Orbis Studio deployment consists of four layers.
Components
| Component | Purpose |
|---|---|
| Nginx | Ingress — routes external traffic, handles TLS termination |
| Orbis Studio | Main application — visual canvas, flow execution, API |
| PostgreSQL | Primary database — flows, users, sessions, settings, metrics |
| Infisical | Secrets management — stores and injects credentials at runtime |
| Langfuse | Observability — traces every flow and LLM call |
| Milvus | Vector database — powers RAG and knowledge retrieval flows |
| MinIO | Object storage — stores Milvus indexes and uploaded files |
Network ports
User access (inbound)
| Port | Description |
|---|---|
| 443 / 80 | Primary user access via Nginx. HTTPS is required in production. |
Management ports (internal network only)
| Port | Service |
|---|---|
| 18080 | Infisical |
| 3002 | Langfuse |
| 5433 | PostgreSQL |
| 19530 | Milvus |
| 9001, 9091 | MinIO / Milvus metrics |
Management ports should never be exposed to the public internet.
Outbound connections
Orbis Studio requires outbound access to:
- The configured LLM endpoint (Azure OpenAI, Ollama, vLLM, or other model provider)
- Any internal systems integrated via flow components
- External services used within flows (if applicable)
System requirements
Platform (no GPU required)
| Resource | Minimum |
|---|---|
| RAM | 16 GB |
| CPU | 6 cores |
| Disk | 150 GB SSD |
The standard production setup runs 3 Orbis Studio replicas behind Nginx for high availability and rolling updates.
On-premises LLM inference (optional)
If you want to run LLM inference locally alongside the platform, a dedicated GPU server is required:
| Resource | Minimum |
|---|---|
| GPU | 1 × NVIDIA A100 80 GB or 1 × H100 80 GB |
Logs
Orbis Studio logs every flow execution and agent step automatically. Logs are stored in the PostgreSQL database and are accessible from the platform UI under Monitor.
Log retention period is configurable by your Orbis Studio administrator based on your organization's requirements. Because the platform runs on-premises, all log data stays within your infrastructure.
For LLM-level tracing (token counts, latencies, per-step breakdowns), see Langfuse integration.