Write safety
Intabula's write path is what makes agent-scale output trustable. A coding agent pointed at your files gets a shell; Intabula's agent gets a toolbox where every operation is designed to be safe: byte-exact edits, atomic multi-file renames under a journal, and a full log of what changed.
Edits splice, they don't rewrite
Changing one field with update_record touches only that segment of the file. Comments, key order, blank lines, and line endings survive byte-for-byte. Most tools round-trip the whole file through a parser and quietly reformat it; Intabula does not.
Renames are atomic and journaled
rename_record is not find-and-replace: one operation rewrites every wikilink across the vault that points at the record. The work is written to a journal first, so an operation interrupted mid-way — a crash, a force-quit — is recovered cleanly on the next launch instead of leaving the vault half-migrated.
Approval mode shows real bytes
With approval mode on, each agent write pauses for your review. The diff is generated by dry-running the actual write without applying it, so what you approve is exactly what lands — not an LLM's description of what it intends. It's off by default, because nobody wants to click approve forty times; turn it on when you want to see changes coming.
Everything is logged
Every agent mutation lands in the activity feed with a before-and-after snapshot of the file, viewable as a diff. Combined with the fact that your vault is plain files — trivially backed up, synced, or put under Git — you always know what changed and can get back to any earlier state of a note.
Soft enforcement, not gatekeeping
Schema violations and broken links are flagged into the needs-attention inbox rather than blocking the write. The vault never refuses your data; it tells you when something drifted. See Data model.