examples/

directory
v2.111.1 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Sep 15, 2026 License: MIT

README

CortexDB Examples

The examples are organized by the current public architecture:

go run ./examples/01_core
go run ./examples/02_rag
go run ./examples/03_memoryflow
go run ./examples/04_knowledge_graph
go run ./examples/05_graphflow
go run ./examples/06_tools_mcp
go run ./examples/07_importflow
go run ./examples/08_self_knowledge_graph
go run ./examples/09_connector
go run ./examples/10_support_brain
go run ./examples/11_unified_brain
go run ./examples/12_incident_agent
go run ./examples/13_scale_analytics
go run ./examples/14_semantic_rag
go run ./examples/15_cortex_query
go run ./examples/16_ontology
go run ./examples/17_query_source
go run ./examples/18_vector_graph_ontology

01_core

Core embedded vector storage through pkg/cortexdb.

  • Opens a single-file DB
  • Uses Quick() for vector add/search
  • Uses collections for namespacing

Use this when you want the smallest CortexDB integration surface.

02_rag

High-level durable knowledge APIs through pkg/cortexdb.

  • SaveKnowledge
  • no-embedder lexical SearchKnowledge
  • entity/relation metadata for graph-aware retrieval

Use this for RAG and app knowledge storage.

03_memoryflow

Agent memory workflow through pkg/memoryflow.

  • transcript ingest
  • recall
  • layered wake-up context
  • diary write/read
  • transcript reconstruction
  • optional Hindsight recall strategy plugin

Use this for chat/session/agent memory.

04_knowledge_graph

Embedded RDF/KG workflow through pkg/cortexdb and pkg/graph.

  • RDF triples
  • SPARQL property paths and subqueries
  • incremental RDFS inference
  • SHACL-lite validation

Use this when you need raw knowledge graph semantics.

05_graphflow

Corpus-to-graph workflow through pkg/graphflow.

  • canonical extraction schema
  • deterministic build/analyze/report
  • graph.json, GRAPH_REPORT.md, and HTML export

Use this when you need to turn documents or model extraction output into an inspectable graph.

06_tools_mcp

Toolbox/MCP-aligned APIs through db.GraphRAGTools().

  • lists the available tool definitions
  • calls tools in-process with JSON payloads
  • demonstrates optional lexical semantic-router tool selection
  • shows the same shapes that can be exposed over MCP

Use this for agents and external LLM orchestration.

07_importflow

Structured-data import through pkg/importflow.

  • parses a CSV (or MySQL/PostgreSQL dump) source
  • builds a MappingPlan routing columns to RAG and knowledge-graph sinks
  • Plan/Run/AutoImport flow with optional AI-assisted mapping

Use this to bootstrap a CortexDB file from existing structured data.

08_self_knowledge_graph

Dogfooding demo: CortexDB maps itself.

  • splits docs/PROJECT_OVERVIEW.md into sections
  • extracts entities/relations with an LLM (local Ollama or any OpenAI-compatible endpoint via EXTRACT_BASE_URL/EXTRACT_API_KEY)
  • builds, analyzes, and exports the graph (JSON, report, HTML viz)
  • qa_test.go answers natural-language questions deterministically from graph edges

Use this as the end-to-end graphflow reference.

09_connector

Privacy-preserving import through pkg/connector.

  • classifies and masks sensitive CSV data
  • imports the safe view into RAG / KG sinks
  • demonstrates the reversible token vault boundary

Use this before connecting live operational data to an agent memory layer.

10_support_brain

Live support-brain workflow.

  • connects to Postgres or MySQL
  • desensitizes rows before RAG/KG ingestion
  • answers with masked evidence and optional unmasking

Use this for a support/copilot memory prototype over live business data.

11_unified_brain

Multi-source unified brain.

  • combines Postgres and MySQL sources
  • streams CDC into one knowledge graph
  • demonstrates SPARQL aggregates, RDFS inference, and SHACL checks

Use this when the agent needs one view across several operational systems.

12_incident_agent

LLM-assisted incident analysis.

  • extracts a KG from unstructured incident reports
  • exposes deterministic search / graph tools
  • lets an agent decide which tool to call before answering

Use this for tool-using agents over unstructured operational text.

13_scale_analytics

Larger-volume analytics harness.

  • timed bulk ingest
  • CDC under load
  • graph analytics and validation over larger fixtures

Use this when you want a rough performance and scale sanity check.

14_semantic_rag

Embedding-backed semantic RAG.

  • plugs in an OpenAI-compatible embedding model
  • demonstrates paraphrase search by meaning
  • uses an LLM answer grounded on retrieved context

Use this when you need true semantic vector retrieval.

15_cortex_query

Composable Cortex Query API.

  • runs dense vector, lexical FTS5, and graph/entity prefetches
  • fuses candidates with weighted RRF
  • shows payload filters, formula boosts, and source-rank debugging
  • runs fully offline with one local CortexDB file

Use this when an agent needs vector + keyword + graph retrieval without a separate vector database or graph database service.

16_ontology

Palantir-style ontology, end to end and fully offline.

  • activates a schema with interfaces, typed object types and per-side link cardinality
  • writes through action types with submission criteria, validate_only and returned edits
  • composes object sets: interface base, filter, intersect, subtract, search_around
  • generates typed MCP tool definitions from the schema (capped, and not auto-registered)
  • diffs a candidate schema and flags the changes that would invalidate stored data
  • closes the generic write path with strict_actions

Use this when you want typed, governed, auditable writes rather than free-form entity upserts.

17_query_source

An external search engine as one retrieval lane, without becoming the storage.

  • implements cortexdb.QuerySource against Meilisearch with nothing but net/http
  • registers it with WithQuerySource and asks for it as a source prefetch
  • fuses it with CortexDB's own lexical lane; every result names the lanes that found it
  • shows what the lane actually buys: a typo (rollbak) that FTS5 correctly cannot match
  • shows the limit: after a row is deleted here, the stale external index still names it and CortexDB drops it — a source names candidates, it never supplies content

Needs a Meilisearch (the file's header has the one-line docker run). Use this when a team already has a search cluster whose recall you want, and you are deciding whether that means replacing CortexDB's storage. It does not.

Rule of Thumb

Need raw vectors / collections?       -> 01_core
Need RAG knowledge storage/search?    -> 02_rag
Need chat/session memory?             -> 03_memoryflow
Need RDF/SPARQL/RDFS/SHACL?           -> 04_knowledge_graph
Need corpus-to-graph/report/export?   -> 05_graphflow
Need agent tool/MCP integration?      -> 06_tools_mcp
Need to import CSV / SQL dumps?       -> 07_importflow
Want the full graphflow E2E demo?     -> 08_self_knowledge_graph
Need privacy-gated live data import?  -> 09_connector / 10_support_brain / 11_unified_brain
Need an LLM tool-using graph agent?    -> 12_incident_agent
Need scale/CDC analytics checks?       -> 13_scale_analytics
Need real semantic vector retrieval?   -> 14_semantic_rag
Need vector + lexical + graph fusion?  -> 15_cortex_query
Need typed, governed, audited writes?   -> 16_ontology
Have a search cluster to fuse in?       -> 17_query_source

18_vector_graph_ontology

All three layers over one estate, and the seams between them.

  • an ontology whose link cardinality refuses the two-writer state at the write
  • object-set algebra: filter, interface, search_around, intersect
  • retrieval joined to the graph through chunk mention edges, in both directions
  • retrieval as one operand of a typed set expression

Runs lexically with no model; set OPENAI_API_KEY to rank the same corpus by meaning. See its own README for the three API defects it surfaced.

Directories

Path Synopsis
Dogfooding demo: turn docs/PROJECT_OVERVIEW.md (CortexDB's own project overview) into a knowledge graph using CortexDB's graphflow workflow, with an LLM extractor.
 Dogfooding demo: turn docs/PROJECT_OVERVIEW.md (CortexDB's own project overview) into a knowledge graph using CortexDB's graphflow workflow, with an LLM extractor.
Demo: desensitize a CSV through the connector privacy gate, then import to RAG.
 Demo: desensitize a CSV through the connector privacy gate, then import to RAG.
Customer-support "agent brain" — an end-to-end CortexDB demo over a REAL database (Postgres or MySQL).
 Customer-support "agent brain" — an end-to-end CortexDB demo over a REAL database (Postgres or MySQL).
Unified support brain — a complex, multi-source CortexDB application.
 Unified support brain — a complex, multi-source CortexDB application.
Incident-analysis agent — a complex CortexDB example that REQUIRES an LLM (a chat/generation model, not an embedding model).
 Incident-analysis agent — a complex CortexDB example that REQUIRES an LLM (a chat/generation model, not an embedding model).
Scale + analytics — a comprehensive, larger-volume CortexDB example.
 Scale + analytics — a comprehensive, larger-volume CortexDB example.
Semantic RAG — CortexDB with a real embedding model (vector search), plus an LLM answer.
 Semantic RAG — CortexDB with a real embedding model (vector search), plus an LLM answer.
Cortex Query — composable retrieval over one local CortexDB file.
 Cortex Query — composable retrieval over one local CortexDB file.
Command ontology demonstrates the Palantir-style ontology end to end: typed object types with mandatory primary keys, link types with per-side cardinality, interfaces for polymorphic retrieval, the composable object set algebra, governed writes through action types, typed tool generation, schema diffing, and the strict_actions write gate.
 Command ontology demonstrates the Palantir-style ontology end to end: typed object types with mandatory primary keys, link types with per-side cardinality, interfaces for polymorphic retrieval, the composable object set algebra, governed writes through action types, typed tool generation, schema diffing, and the strict_actions write gate.
Command query_source shows an external search engine acting as one retrieval lane inside CortexDB, without becoming CortexDB's storage.
 Command query_source shows an external search engine acting as one retrieval lane inside CortexDB, without becoming CortexDB's storage.
Vectors, a knowledge graph, and an ontology — one brain, one estate, and what each layer can answer that the other two cannot.
 Vectors, a knowledge graph, and an ontology — one brain, one estate, and what each layer can answer that the other two cannot.
Command kg_e2e is a runnable, fully-printed end-to-end walkthrough of the CortexDB RDF / Knowledge Graph stack through the public pkg/cortexdb facade.
 Command kg_e2e is a runnable, fully-printed end-to-end walkthrough of the CortexDB RDF / Knowledge Graph stack through the public pkg/cortexdb facade.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.