AgenticData v0.1.0

Universal data comprehension for AI agents — infer schemas, track lineage, detect anomalies, transform any format.

Highlights

• 16 format parsers with auto-detection (CSV, JSON, XML, YAML, HTML, SQL, Log, Email, Calendar, GeoJSON, KML, GPX, Markdown, and more)
• 19 semantic field types — infers Email, URL, IP, GeoPoint, Currency, DateTime from raw sample values
• 122 MCP tools across 24 capability groups
• Field-level lineage with trust scoring (Equation: trust(d) = max(0.1, 1.0 - 0.05d))
• 5-dimensional quality scoring — completeness, uniqueness, consistency, freshness, accuracy
• PII detection — email, phone, SSN, credit card (Luhn-validated), IP address with 4 redaction methods
• Field-level encryption with BLAKE3 per-field key derivation
• .adat binary format — LZ4 compressed, BLAKE3 checksums, 136.5 bytes/record
• Graph engine — BFS traversal, degree centrality, automatic FK discovery
• Session management — per-user operation tracking with resume
• 265 tests — all passing (0 failures)

Install

curl -fsSL https://agentralabs.tech/install/agentic-data | bash

Binary Contents

• adat — CLI with 8 commands (ingest, detect, query, quality, pii, status, formats, schema)
• agentic-data-mcp — MCP server with 122 tools

Research Paper

Paper I: AgenticData — Universal Data Comprehension (PDF)

Full Changelog: https://github.com/agentralabs/agentic-data/commits/v0.1.0