Predictable Evidence
Runs, stages, metrics, events, artifacts, and reports share one local evidence model.
Local-first observability for machine learning systems. Capture enough structured evidence that a run can be reviewed, compared, diagnosed, reported, and recovered later.
runrun:wine.random-forest
stageevaluate
metricaccuracy = 0.96
artifactmodel.txt
reportsnapshot-report.md
Try it locally
Install Contexta, then use the copy-and-run ML, DL, or LLM example in Getting Started to capture local evidence.
uv add contexta
Why Contexta
Local-First
Start in a notebook, CI job, or worktree without a hosted service or collector process.
Reviewable Workflows
Compare candidates, inspect diagnostics, trace lineage, and build reports from the same workspace.
Recoverable State
Backup, verify, restore, replay, and transfer workflows are part of the product surface.