Achilles — Knowledge Store

A multi-model store: a single knowledge entity lives as three projections in one Postgres — a relational body, chunk vectors, a graph of relations. This is the platform's retrieval layer: Harvester writes here, Query Engine draws context from here for RAG. The distinctive trait — four search paradigms fused under a single ACL in one SQL query.

Architecture

Producerstier 1 · surface

Harvester

writes: content · structural edges · access rights · people

writes ▼

Storage · one PostgreSQLtier 2 · core

entityknowledge node · source of truth

entities

relational body

exact filters and joins

chunks

chunk text

vector pgvector·HNSW + lexical tsvector·GIN

entity_edge

graph of relations

recursive SQL·CTE · no Neo4j

Retrievaltier 3 · fusion

one ACL pre-filter · identity at the entry narrows candidates before ranking · a single JOIN for all 4 primitives

vector

semantics

lexical

words

graph

relations · CTE

sql

exact facts

▼

RRF fusionfusion key = entity

reads ▲

Consumerstier 4 · output

Query Engine

RAG over knowledge · owns the dialogue

Agent Engine

4 primitives as tools

API

POST/api/v1/retrieval/vector · lexical · graph · sql
POST/api/v1/retrieval/hybrid
GET/api/v1/admin/knowledge/sources
POST/api/v1/admin/knowledge/reindex
POST/api/v1/admin/knowledge/re-embed
POST/api/v1/admin/knowledge/backup
POST/api/v1/admin/knowledge/restore

Curation, re-embed and backup status reach open screens as push nudges over the platform live-updates stream; the endpoints above stay the single source the client refetches.

Workzone Design

Detailed design decisions for the module.

Data model Hybrid search ACL and identities Lifecycle Stack Embeddings runtime Tests