OrionBelt®: Semantic Sidecar for Agentic AI & Governed Analytics

OrionBelt® is a source-available platform of five tools that lets AI agents go from a raw database schema to compiled, dialect-safe analytical SQL in a single conversation, and to scheduled, audit-ready reports on a cron. At its core, the OrionBelt® Semantic Layer (OBSL) is a Semantic Sidecar for AI, analytics, and governed data systems (orionbelt.ralforion.com): it injects governed business semantics into existing platforms with no architecture change. Agents query in business concepts, never raw schemas. The platform combines ontology generation, a YAML-based semantic layer, an OWL/SKOS ontology workbench, a scheduled-report runner, and a conversational AI interface, all connected via MCP (Model Context Protocol).

The Platform

OrionBelt® Analytics

MCP server for Text-to-SQL on the data you already have. Generates an RDF/OWL ontology of your schema; GraphRAG finds the tables and join paths a question needs, even without declared keys, and inferred joins can be checked against the data; OBQC checks every query and blocks fan-traps before they run. Read-only, several named databases per server, credentials never in the chat, your own ontology per session. Not a BI tool and not an LLM: charts appear in the chat, you bring the model. Docker image ships with English and multilingual embedding models. Docs

MCP RDF/OWL GraphRAG OBQC Text-to-SQL 30 tools

OrionBelt® Semantic Layer

Source-available Semantic Sidecar: injects governed business semantics into existing AI, analytics, and data platforms with no architecture change. Compiles YAML models (OBML) into optimized SQL across 8 dialects via custom AST. Agents query in business concepts, never raw schemas. Freshness-driven cache, agent-facing API.

Live: orionbelt.ralforion.com

Semantic Sidecar Analytics as Code AI Governance REST API 8 Dialects MCP Apache Arrow Flight

OrionBelt® Runner

Run OBML query batches against the Semantic Layer and emit reports. YAML-defined runs produce self-contained MD/HTML reports plus an audit-grade YAML run-log with compiled SQL, query plans, and timing.

YAML Runs MD/HTML Audit Trail Cron/CI

OrionBelt® Ontology Builder

Browser-based OWL & SKOS workbench. Streamlit + rdflib. No Java, no Protégé. Bulk operations, a read-only SPARQL console, OWL-RL reasoning, 22 SKOS checks with one-click fixes, gist upper-ontology starters, merge-aware imports.

Live: orionbelt.streamlit.app

OWL SKOS SPARQL Streamlit OWL-RL

OrionBelt® Chat

Conversational AI tying it all together. 300+ models via OpenRouter; Anthropic/OpenAI direct; local via MLX or Ollama. Dual MCP servers, MCP sampling, Plotly & Mermaid inline.

Chainlit Pydantic AI Multi-LLM MCP Sampling

Database Schema

→

Analytics (Ontology)

→

Semantic Layer (OBML)

→

Compiled SQL

→

Runner (Reports)

→

Charts & Insights

Key Differentiators

Semantic Sidecar pattern: drop-in governed semantics for existing AI, analytics, and data systems; no architecture change, no dedicated semantic infrastructure
Analytics as Code: version-controlled YAML compiled into executable SQL, DQ rules, KPIs, and semantic context; one model powers AI agents, analytics, DQ checks, and reporting
Governed AI data access: agents query the semantic layer, not raw tables; consistent metrics, no SQL hallucinations, full audit trail
AST-based SQL generation: custom SQL AST ensures correct, injection-safe SQL (not string templating)
Agent-facing API: model-health on load, query-plan endpoint, structured warning codes; freshness-driven result cache with heartbeat invalidation
Checked Text-to-SQL (OBQC): OrionBelt® Analytics validates every generated query against the ontology and blocks fan-traps, unknown columns, and missing joins before execution
Multi-fact queries (CFL): Composite Fact Layer handles queries spanning independent fact tables via UNION ALL
MCP-native: works with Claude, ChatGPT, Copilot, Cursor, Windsurf via stdio or hosted Streamable HTTP
Scheduled reports & audit trail: OrionBelt® Runner emits MD/HTML/PDF reports with a YAML run-log capturing compiled SQL, plans, and timing
OSI interoperable: bidirectional with Open Semantic Interchange; OBSL graph + SPARQL on every loaded model

Supported Databases

PostgreSQL Snowflake BigQuery ClickHouse Databricks DuckDB Dremio MySQL

Architecture

REST API (FastAPI) with OpenAPI docs
MCP Server for AI agent integration
PostgreSQL Wire Protocol endpoint (psql, JDBC, ODBC, BI tools)
Gradio UI for interactive exploration
DB-API 2.0 + Apache Arrow Flight SQL drivers
OBSL Graph + SPARQL querying
Docker images on Docker Hub
Hosted demo on Google Cloud Run (HTTPS, wildcard cert, scale-to-zero)

Live Endpoints: orionbelt.ralforion.com

API: /v1/*, REST + /docs (Swagger), /redoc
UI: /ui/, Gradio playground with pre-loaded example model
MCP: /mcp, Streamable HTTP, no auth, paste the URL into your MCP client

Use Cases

AI-Assisted Analytics

AI agents query your data through semantic models instead of writing raw SQL. Guaranteed correct joins, aggregations, and dialect-safe output.

Multi-Cloud Data Access

One semantic model, eight SQL dialects. Write once, deploy to Snowflake, BigQuery, Databricks, or any supported warehouse.

Governed AI Data Access

The semantic layer is the gate AI agents go through to reach your data. Consistent metrics, RDF lineage, SPARQL querying, no raw-SQL hallucinations.

OrionBelt®: Semantic Sidecar for Agentic AI