Basalt

Fetch the currently deployed version of a prompt by its slug, ready to pass to any LLM call in your pipeline.

Retrieve all prompts in your Basalt workspace — names, slugs, and current deployment status — for inventory and governance tasks.

Retrieve full metadata for a specific prompt — description, all versions, deployment tags, and edit history — for documentation and change management.

Assign or update a deployment tag on a specific prompt version — promoting a tested variant to 'production' or rolling back to a previous version.

List all datasets in your Basalt workspace, including slugs and metadata, for dataset governance and experiment planning.

Retrieve a specific dataset by slug along with all its rows — for inspection, reporting, and feeding examples into LLM evaluations.

Append a new labelled example row to a Basalt dataset — capturing live production inputs and outputs for ongoing evaluation and regression testing.

Create a new experiment in Basalt to compare prompt variants, model choices, or pipeline configurations — tracking runs and evaluation scores in one place.

Log a single prompt input/output pair to Basalt monitoring — capturing the model, prompt version, latency, and any evaluation scores for production observability.

Start an OpenTelemetry-compatible trace in Basalt to instrument a multi-step AI workflow — capturing every LLM call, retrieval step, and tool invocation end-to-end.

Retrieve the version history of a prompt to track how it evolved, identify when a regression was introduced, and select the right version to roll back to.

Retrieve traces from Basalt observability to analyze AI workflow performance, surface latency outliers, and identify failing steps across production runs.