General
generalRequired fields, logical types, nulls, unique IDs, ranges, formats, supplied references and freshness.
LYNVERA OPEN SOURCE
Lynvera AIDataReady
Know your data.
Build reliable AI.
Check the data before you build on it. Evaluate explicit contracts for AI applications with a local Python library and CLI.
Source version 0.1.0 / Python 3.12+
Local executionNo model requiredNo telemetry
FOUR PROFILES
Choose the contract your data needs.
Start with shared checks. Add rules for how your application uses the data.
RAG
ragDocument IDs, usable text, provenance, versions, exact duplicates and required filter or permission metadata.
Machine learning
machine-learningConfigured targets, class balance, cardinality, temporal splits and explicitly declared leakage risks.
Agent tools
agent-toolsTool fields, documented schemas, contract versions, allowed actions and authorization metadata.
SOURCES
Work with the data you already have.
CSV
Tabular files
JSON
Arrays of records
JSONL
One record per line
Parquet
Columnar files
PostgreSQL
Read-only access
In memory
Python records
PostgreSQL uses allowed tables, bound equality filters, read-only transactions and timeouts. Credentials stay in an environment variable.
PostgreSQL setupA REPEATABLE WORKFLOW
Define. Evaluate. Review.
- 01
Define a contract
Choose a source, a profile and the fields and thresholds your application requires.
- 02
Run the assessment
Use the CLI or Python API to check your data locally against that YAML contract.
- 03
Review the findings
Inspect status, severity and violation counts. Keep reports with your development or CI evidence.
GET STARTED
Your first assessment, locally.
Version 0.1.0 is available from source. Use Git and Python 3.12 or newer. These commands pin the reviewed examples and documentation.
git clone https://github.com/lynvera-ai/aidataready.git
cd aidataready
git checkout a7c2c4b3e429617ccb4ad293209e09f2b8e519b6
python3 -m venv .venv
source .venv/bin/activate
python -m pip install .For Windows, uv and optional dependencies, follow the installation guide. Installation guide
aidataready check --config examples/rag.yaml --output rag-result.json
aidataready report --input rag-result.json --format html --output rag-report.htmlfrom lynvera_aidataready import Assessment
result = Assessment.from_config("examples/rag.yaml").run()
print(result.status)
print(result.rows_examined)READ THE RESULT
A passing contract. A known failure.
These are synthetic repository fixtures, not customer data or a live assessment. Both contain two records.
PASSExit code 0
RAG documents
The supplied text and metadata satisfy the configured checks.
Verified CLI summary
local-knowledge: PASS. Records: 2.profile: rag / rows_examined: 2
FAILExit code 1
Duplicate IDs
Two records share one ID. The uniqueness check reports one violation.
Verified CLI summary
duplicate-ids: FAIL. Records: 2.structure.unique / id / violations: 1
aidataready check --config examples/duplicate-ids.yaml --output duplicates.jsonFAIL exits 1. Run report conversion separately after that expected failure.
PASS and WARN exit 0. FAIL exits 1. ERROR and INCONCLUSIVE exit 2. Empty or partial evidence cannot establish a passing scan.
Examples verified November 1, 2026. The RAG fixture uses fixed dates and a 3,650-day freshness window. Verification record
REPORTS
Keep evidence in the format you use.
JSON
Structured findings and aggregate statistics. Preserves each original status.
Download synthetic example JSONHTML
A local report with severity, category and source filters. No external assets.
Download synthetic example HTMLJUnit
XML for CI. WARN and INCONCLUSIVE map to skipped tests, so retain JSON when those distinctions matter.
Download synthetic example JUnitaidataready report --input rag-result.json --format json --output report.json
aidataready report --input rag-result.json --format html --output report.html
aidataready report --input rag-result.json --format junit --output report.xmlReport conversion does not rescan data. Reports contain counts and field names, without record samples.
DOCUMENTATION
Go from a first check to your own contract.
Questions, answered.
Does PASS certify that my AI will perform well?
No. Assessments check the supplied data contract. They do not certify model performance, retrieval relevance, semantic quality or effective access permissions.
Does my data leave my machine?
Local file and in-memory assessments do not upload your data. The package sends no telemetry. PostgreSQL connects to the database you configure. Installing dependencies usually needs network access.
Is sensitive data detection definitive?
No. Optional sensitive data detection is heuristic. It can miss sensitive data or flag ordinary values. Review the findings and your own privacy requirements.
Are batch budgets absolute memory limits?
No. Batch budgets are checked after decoding and are not hard process memory caps. Distinct tracking is also bounded. Incomplete evidence may produce INCONCLUSIVE.
Do I need a model, account or paid service?
No. The synthetic examples run locally after installation. PostgreSQL requires your own database and its optional driver.
Can I install the package from PyPI?
The reviewed 0.1.0 project is distributed from source. Use the source installation shown here and check the repository release guidance before upgrading.
Make the contract explicit.
Start with the included data. Then define what your own application needs.