How Inherent Works
A deep dive into the architecture that powers production-grade RAG infrastructure
Architecture Overview
Inherent is built on a strict architectural separation between Truth and Memory:
Postgres
Contextual Truth Layer
- Documents
- Versions
- Chunks
- Metadata and lineage
Weaviate
Contextual Memory Layer
- Embeddings
- Hybrid semantic search
Ingestion Flow
Content Ingestion
Connectors send content to the /ingest API endpoint. Content can come from Git repositories, Wikis (Confluence, Notion), file systems, APIs, or custom connectors.
Normalization & Chunking
Content is normalized to a consistent format and intelligently chunked to preserve semantic meaning.
Version Storage
Document versions, chunks, and metadata are stored in Postgres (the truth layer). Every change is versioned—no silent overwrites.
Semantic Indexing
Embeddings are generated and indexed in Weaviate (the memory layer) for fast semantic search.
Context Retrieval Flow
Search Request
AI applications call the /search API with a query.
Semantic Search
Weaviate performs hybrid semantic + metadata search to find relevant candidate chunks.
Truth Hydration
Postgres hydrates the authoritative truth for each candidate, ensuring no stale or orphaned chunks.
Context Pack Assembly
A deterministic context pack is returned—source-aware, version-aware, metadata-rich, and structured. Exactly what serious AI systems require.
Key Architectural Principles
Separation of Concerns
Truth (Postgres) and Memory (Weaviate) are strictly separated. This ensures auditability, rebuildability, and prevents cross-contamination.
Versioning by Default
Every document and chunk is versioned. You can always rebuild the system from source and trace the history of any piece of knowledge.
Deterministic Retrieval
Context retrieval is deterministic. The same query returns the same context pack, guaranteed by the truth layer.
Shared Backbone
All AI applications pull from the same trusted system. No duplication, no inconsistency, no fragmentation.
Ready to get started?
Start building production-grade AI applications with Inherent.