Changelog¶

All notable changes to this project are documented here. The format follows Keep a Changelog and the project adheres to Semantic Versioning.

While the version is 0.x, minor releases may contain breaking changes. They are always listed under Changed or Removed.

Unreleased ¶

agentl.pandas: @dataframe turns fn(df) -> DataFrame into a step, plus to_df and to_records; agentl-etl[pandas] extra (#8).
postgres.read and postgres.write (append, replace, upsert on key), with typed auto-created tables and batched writes in one transaction; agentl-etl[postgres] extra (#9).
Readers accept url: (HTTP download with headers and retries) besides path:, and upstream records whose path is a URL (#7).
file.read: detects the format by extension, expands ZIP and GZ archives safely (path traversal rejected, size cap) and reads mixed folders (#22).
New readers: txt.read, parquet.read, doc.read (via LibreOffice), xml.read, html.read; excel.read now also reads legacy .xls. New extras: xls, parquet.

The PyPI distribution is now agentl-etl (PyPI treats agentl as confusable with an existing project). Install with pip install agentl-etl; the import (import agentl) and the agentl command are unchanged. 0.1.0 was never published.

YAML pipeline spec with ${VAR} interpolation and extract/transform/load sections.
Runner that executes steps in graph order and emits run, step and item events.
Agent steps with schema-validated output, per-record cache and rejected items.
Connectors: CSV, JSON/JSONL, Excel, PDF, Word, HTTP, SQLite, select, filter, dedupe, join, concat.
LLM providers: Anthropic and an offline echo provider.
Notifications to Slack, generic webhooks and the console.
CLI (run, validate, diagram, runs, connectors, dashboard) and a local dashboard.