Undo any change to any table.
Every change to a Spiral table is a commit. Branch to test a change. Tag the version that a model trained on. Read any table as of a commit, tag, or branch.
Old versions stay readable as open Vortex files.
Agents make mistakes at machine speed.
Branches, snapshots, commits, permissions, and provenance make shared data safe for a team of people. Agents need them more, because an agent runs many experiments at once, each on its own branch.
Branches
Run many experiments and merge one. A branch makes no copy of the data.
Snapshots
Read any table as it was at a commit or tag, or roll the table back to it.
Multi-table commits
Each project has a serializable transaction log. A pipeline changes all of its tables or none of them.
Project roles
Roles decide who can read and who can write each project. An agent signs in as its own workload, with its own role. An app that acts for a person gets no permission that the person lacks.
A training set is a filter and a tag.
Curation copies nothing. A training run reads the training set at its tag and records the commit behind it, so the team knows which rows the model saw.
Each change is a commit that a person can review or undo.
Yes: Runs pinned to a tag
Each run record names the exact table versions behind the tag, the mix weights, the seed, and the batch size.
No: Folders named _final_v3
_final_v3 becomes a tag on a commit, and nobody has to guess what “final” meant.
Yes: Agents on their own branches
An agent writes to its branch. A person reads the diff and merges it.
No: An incident when an agent deletes data
In July 2025, a coding agent deleted a production database during a code freeze. In Spiral, reverting the agent’s commit restores the table in one step.
What it builds on.
Workflows:Investigate,Enrich,Curate,and Train