datatally

An AI model pays for data in tokens and latency, not in dollars, yet a dataset byte count or row length tells you nothing about whether it is worth a single inference call. DataTally addresses this gap by aggregating public usage signals of data assets — downloads, citation counts, forks, commits, stars — into a single verifiable profile that carries its source and fetch timestamp on every metric, giving AI an objective basis for choosing data instead of relying on subjective labels.

The plugin exposes three tools: search_assets to look up profiles by keyword or domain, get_asset_profile to pull a full per-source metric timeline for one asset, and compare_assets to contrast two assets strictly within the same signal layer. It draws a clear line between deep signals (actual use: downloads, citations, forks) and shallow signals (interest: stars, likes), and never treats one layer as inherently stronger than the other. Every missing field is surfaced as null, single-source assets are explicitly marked, and no default value is ever fabricated. DataTally does not compute value, rank assets, or recommend one over another; the recording is its job, the judgment is yours.

If you are building an AI agent, a researcher who needs a model to autonomously pick the right dataset, or an engineer orchestrating data workflows where verifiable usage evidence should replace subjective review, DataTally plugs in as a lightweight memory layer that lets your agent inspect real usage signals before it ever invokes a dataset.