CoreCortex Core · developer preview

Retrieval systems
you can test.

CoreCortex helps you turn a collection of documents and data into a searchable corpus. It records how the data was prepared, runs retrieval tests, and shows which configuration works best.

CORPUSevidenceversion 24
sourcerecords12,482
strategyhybridcandidate 06
evaluation0.842+ 4.8%
For corpora where the source matters
CORPUSRETRIEVALEVALUATIONGOVERNANCE
How it works

Every retrieval result depends on choices you should be able to inspect.

CoreCortex records the source data, the passages made from it, the search index, and the test results. When a result is poor, you can see which choice to revisit.

01

Inspect

List the source files, look at real records, and write down the questions people need to answer.

02

Prepare

Give each record a stable ID, keep its source details, and split long documents into searchable passages.

03

Search

Try keyword, vector, filtered, relational, or graph search where each one fits the data.

04

Test

Run reviewed questions, inspect the misses, and keep the configuration that performs best.

One configuration file

Describe the corpus and how people will search it.

The corpus already exists in your documents and systems. Thecorec.yaml file tells CoreCortex where the data comes from, how to prepare it, which indexes to build, and which retrieval plan to use.

Read the configuration guide →
corec.yamlv1alpha1
apiVersion: corecortex.io/v1alpha1
kind: Corpus
metadata:
  name: knowledge-operations
spec:
  datasets:
    - name: reference-guides
    - name: change-records
  indexes:
    - name: evidence-search
  retrievalPlans:
    - name: evidence-investigation
✓Valid configuration · 2 datasets · 1 retrieval plan
Core and Enterprise

Start with Core. Add automation when you need it.

Core provides the data model, command line, adapters, and evaluation formats. Enterprise adds tools for onboarding data and running many controlled experiments.

OPEN SOURCEDeveloper preview

CoreCortex Core

Build and test retrieval with the open-source command line and APIs.

  • ✓Declarative corpus configuration
  • ✓Canonical evidence and provenance
  • ✓Normalized record ingestion
  • ✓Pluggable storage interfaces
  • ✓Inspectable retrieval plans
  • ✓Evaluation harnesses
View open-source scope ↗
ENTERPRISEPrivate preview

CoreCortex Enterprise

Prepare customer data and automate repeatable retrieval experiments.

  • +Maintained ingestion connectors
  • +Schema and storage discovery
  • +Generated evaluation candidates
  • +Continuous strategy experiments
  • +Promotion gates and monitoring
  • +Enterprise operations
Become a design partner →
Retrieval experiments

Change one setting and measure the result.

Compare document conversion first, then chunking, embedding models, search methods, and the number of results sent to an answer model. Use the same reviewed questions for every comparable run.

qualityDoes it find the right evidence?
latencyCan it answer on time?
costIs the strategy sustainable?
accessDid it hide restricted records?
Explore CoreCortex Enterprise →
EXPERIMENT LOGACTIVE DESIGN
041
baseline · hybrid-v4nDCG@10 · 0.803
BASE
042
section-aware chunksnDCG@10 · 0.831
KEEP
043
graph expansion · depth 2nDCG@10 · 0.824
DROP
044
citation relation boostnDCG@10 · 0.842
BEST
changebuildevaluatedecide
Public documentation

Start with a task and follow it through.

Browse all docs →
CoreCortex · source-linked retrieval

Build retrieval you can debug.

Start with real records and real questions. Keep the source attached to every result. Measure each change before you keep it.

Read the docs →Talk to us