All pages

Docs · IndexMarkdown

Documentation

Metrics by 041 is ML experiment tracking built for research agents and the people who work with them. A training job logs with the Python collector; agents and scripts read back with the command line, DuckDB or the Python API client; people look at charts in the app. These pages say what is true for someone doing that. Each page is short and answers one question; follow the links for the rest.

Every page is plain markdown. The site serves the same text at /docs/<page>.md, answers any page with markdown for curl or Accept: text/markdown, and /llms.txt lists every page.

Start here

what experiment tracking needs when agents run the research, and what we left out

sign in, log a first run from Python and read it back, in five minutes

Working as an agent

the command line, keys, markdown docs and the habits that keep results honest

organizations, folders, experiments, metrics, metadata and annotations

Log

syvain-metrics-collector: the Python API, its arguments and what it guarantees

names, metadata, cadence and the field every value belongs in

Read and analyze

syvain-metrics: sign-in, folders, experiments, series and charts as JSON

the syvain_metrics extension: tables, selectors, the cache and example queries

Python API client

syvain-metrics-api-client: folders, experiments and series from scripts

The app

the folder tree, the loss chart, diagnostics, chart controls and saved views

Organizations

Organizations and keys

members, invitations, roles and API keys, from the app or the CLI

Limits and fair use

every limit the API enforces, what free means, and how to ask for more

HTTP API

authentication, event envelopes, batches, errors, retries and streaming series

every v2 event with its data and reply, generated from the API contracts

Under the hood

Syvain Series DB, the storage we built for experiment metrics: lossless, fast and cheap to run