Qanat

•
331
•

Ingests corporate filings and calculates point-in-time financial metrics for valuation and historical screening without look-ahead bias.

Qanat is an investment research data platform designed to process corporate filings and evaluate financial metrics as of any historical date. By tracking two distinct dates for every figure—the period covered and the actual publication date—Qanat eliminates look-ahead bias in financial screening and backtesting. Built on a Parquet and DuckDB architecture for concurrent, multi-user file storage, Qanat exposes its pipelines, documents, and historical screening engine through an MCP interface with granular access scopes, allowing AI agents to safely query company fundamentals, verify filing coverage, and track returns without data leakage.

主要功能

01330 GitHub stars
02Point-in-time historical screening that accounts for both reporting periods and publication dates to eliminate look-ahead bias
03Parquet and DuckDB storage architecture enabling non-blocking, multi-user read access and low storage footprints
04Tiered MCP scoping to control tool exposure for agents, ranging from read-only data access to pipeline authoring
05Built-in ingestion connectors for SEC EDGAR (10-K and 10-Q forms), Yahoo Finance, HTTP, SQL, and CSV sources
06Versioned metric definitions and execution digests for verifiable, reproducible investment research results

使用案例

01Quantitative Backtesting: Running point-in-time stock screens to test investment hypotheses without survivorship or restatement bias.
02AI Financial Research: Equipping LLM agents with the ability to inspect SEC filings, verify historical company fundamentals, and evaluate screen criteria.
03Pipeline & Metric Versioning: Defining and maintaining reproducible financial metrics and transformations over structured corporate disclosures.