Design Archive
A public archive of internal design thinking. These documents capture architecture decisions, DSL design, runtime behaviour, and strategic direction as Osyrin evolves.
architecture
System Overview
High-level architecture of the Osyrin platform: compiler, runtime, and deployment model.
Entity-Anchored RAG
How the platform unifies structured entity data and unstructured file content into a single semantic layer, configured per entity type through a declarative DSL block. Part 1 of 3, covering the data model, the RAG block syntax, the ingestion pipeline, and the evolution from inline per-entity vectors to a consolidated architecture.
RAG Security, Classification, and Knowledge Graphs
Part 3 of 3 on the platform's RAG system. [Part 1](rag-foundations.md) covers the data model, ingestion pipeline, and Pulse synthesis. [Part 2](rag-pulse-and-agents.md) covers Pulse, entity chat, and agent interaction patterns. This document covers how the platform ensures the right people see the right data, how citations create verifiable knowledge, and how soft references build an emergent knowledge graph.
RAG Part 2
The AI-powered layer built on top of entity-anchored RAG foundations. Part 1 covers the data model, context shadows, ingestion pipelines, and embedding infrastructure. This document covers what we do with those foundations: auto-refreshing entity summaries, the universal chat interaction pattern, multi-agent pages, insight extraction, cross-entity search, long-running memory, and workflow agent actions. Part 3 covers security, citations, and knowledge graphs.
MCP Server
How the platform exposes application functionality to AI clients (Claude, Cursor, custom agents) via the Model Context Protocol, with role-based tool visibility, typed schemas, and full RBAC enforcement.
REST API
How the platform exposes application logic and data as REST APIs with OpenAPI documentation, configurable authentication, entity CRUD, and per-consumer access control — all declared in the DSL and compiled to metadata.
Query Engine
How the platform's query engine evolved from a single-pass SQL compiler with known gaps into a multi-phase scope tree architecture supporting aggregates, projections, per-entity security, and a first-class DSL block for defining reusable data views.