---
updatedAt: 2026-08-10T19:28:07.000Z
agentTools:
  projectIndex: https://docs.getsolid.ai/llms.txt
---

# Platform Architecture

How Solid's cloud platform is architected — compute isolation, encryption, and the "no data sent to LLM" boundary.

Solid's SaaS platform runs on Microsoft Azure.

## Compute Isolation

Solid runs on a private, dedicated Kubernetes cluster with compute segregated into two pools:

* **Real-time services** — the MCP server, the SQL generation API, and the Solid Build and Analyze interfaces
* **Data pipelines** — schema ingestion, query log processing, model generation, and benchmark execution

This separation means a heavy ingestion job for one customer's data can't degrade the responsiveness of live MCP/SQL generation traffic for another.

## Data Storage and Encryption

* All operational data (semantic models, benchmark questions, metrics library, user data) is stored in a managed, highly-available database
* All storage components are encrypted at rest using **Customer-Managed Keys (CMK)**
* All traffic is encrypted in transit via **TLS 1.2+**

## AI Boundary — "No Data Sent to LLM"

Solid's LLM services run through Azure AI Foundry (Azure OpenAI, plus document processing, translation, and vector/semantic search for model routing and glossary lookup).

**Critical boundary:** the LLM processes schema metadata, query patterns, and semantic model context — not raw customer data. Raw data never passes through the LLM boundary. See [Security Architecture](https://docs.getsolid.ai/docs/security-architecture) for the full "zero-raw-data" model.

## Observability

Solid maintains operational logging, tracing, and monitoring across the platform to support incident response and audit requirements. Details are available on request as part of a security review.