Changelog
Turn completed work into reusable knowledge
Opt-in background learning can save lessons from your work, with control over what gets reviewed and what gets kept.
Less repeating yourself
A useful correction shouldn’t have to stay buried in a conversation. With automatic learning enabled, Prokop can review eligible completed assistant turns and save reusable lessons to memory and user preferences. You can also allow it to create or refine skills.
This is background review, not model training. Shared workspace knowledge belongs to the project. Personal knowledge belongs to an agent in its home workspace and can help it across projects.
You choose what it learns from
Learning is off by default. Enable Automatic learning in workspace settings to use it. This also enables Memory and Session search, which the review process needs.
Choose your learners, their models, their focus, and their review timing. You can preview the configured prompt before saving. Personal learning can draw from selected eligible workspaces, and each workspace can disallow being used as a personal learning source.
For a conversation you want left out, use Exclude from learning in the session menu. Excluding it stops future learning from that conversation; it does not erase knowledge already saved.
Inspect the changes
Learning history shows the runs, their source conversations, and per-file diffs. If a saved change isn’t useful, you can undo it individually, provided the file still matches that learning result. Later edits are not overwritten.
Also: discard changes to one file
The Changes tab now has Revert changes… in a modified file’s row menu. After confirmation, it restores that tracked file to its last committed version, discarding both staged and unstaged changes to that file. Unrelated files stay untouched.