A project that follows the model
One workspace represents one model lifecycle. Research, immutable dataset versions, recipes, runs, and exports stay connected through model-version lineage.
MODEL DEVELOPMENT / GITHUB
A dedicated workbench for turning research and data into a model you can train, evaluate, and understand.
THE PROJECT, IN PRACTICE
One workspace represents one model lifecycle. Research, immutable dataset versions, recipes, runs, and exports stay connected through model-version lineage.
Work with a typed graph or edit the recipe as JSON. Revisions, bounded iteration, conditions, and artifact ancestry preserve the structure of an experiment.
The workbench and native MCP server use a shared authenticated operations service. External agents can work with the same project operations.
Dataset previews, consumption records, checkpoint evaluation, and functionality-family policies connect training runs to their data and evaluation results.
FROM IDEA TO USE
Start a model project and organize source material, datasets, and the code it depends on.
Choose training and evaluation components, configure resources, and retain a versioned recipe.
Inspect checkpoints and evaluation results, compare model versions, and prepare an export with its lineage intact.
research/ Papers, citations, source material
datasets/ Registered versions and previews
recipe Training + evaluation workflow
runs Checkpoints, metrics, artifacts
model versions Comparisons and lineage
exports Reviewed publication bundlesConceptual workspace map—not a required on-disk directory structure.
TAKE IT FOR A SPIN
Protomire is in development. Start with the project quickstart and the installation-status page. Developers can build from source using the repository runbook.
Development builds exist for macOS and Linux with scoped acceptance. There is no published, signed, clean-machine-accepted release. Windows delivery is unfinished. First-party licensing has not yet been selected.