Skip to content

Persisting

Persistent Infrastructure for the Agent Era

Capture, import, and query Agent trajectories as durable history.

From captured events to Datasets, capture, analysis, and durable history follow one traceable path.

Quick start
$ pip install persisting$ pchronicle onboard

Agent history = Dataset + query

Make agents easier to understand and improve.

A harness keeps an agent working. Chronicle keeps the run as durable, queryable history so the next decision can see what actually happened.

Datasets

Mount captured or imported Sources and see run counts before you drill in.

Trajectory

Reconstruct a complete run from one event stream: prompts, tool calls, results, and context injection.

Analysis

Ask a question or run bounded SQL against the same Snapshot the warehouse is serving.

Design approach

Every run is a Dataset. Every query is scoped.

Facts before interpretation

Preserve source events before building queryable projections.

Every run is traceable

Inspect records by Source; query, resume, and replay operate on the same event stream.

Dataset overviewRun trajectory

Documentation surfaces

Start with the question you have.

Open source · Apache-2.0 · Persisting

GitHub