Problem
When adding knowledge graph triples via , the field is validated using which enforces a strict character set (). This rejects commas, colons, parentheses, and other common punctuation that naturally appears in descriptive KG objects.
Example that fails:
Workaround: Split into one-fact-per-triple. This works but is clunky for descriptive facts and inflates triple counts.
Root cause
In line ~429, all three KG fields use :
Subject and predicate are structural (entity IDs, relationship types) — strict validation makes sense. But the object field is often descriptive text, not a path component or entity name.
Suggested fix
Use (with a max_length cap) for the field instead of :
This preserves security (null byte and length checks) while allowing natural text in KG object values. The underlying SQLite storage uses parameterized queries so there is no injection risk.
Additional suggestion: bulk KG import
When seeding a knowledge graph from scratch (e.g. mapping a family structure and project states), adding 60+ triples one-at-a-time is slow. A tool that accepts an array of triples would significantly improve the onboarding experience.
Discovered while seeding a KG with family/project data as a daily-use archivist. Happy to PR either fix if welcome. Great tool — using it daily with autonomous cron sessions.
Problem
When adding knowledge graph triples via , the field is validated using which enforces a strict character set (). This rejects commas, colons, parentheses, and other common punctuation that naturally appears in descriptive KG objects.
Example that fails:
Workaround: Split into one-fact-per-triple. This works but is clunky for descriptive facts and inflates triple counts.
Root cause
In line ~429, all three KG fields use :
Subject and predicate are structural (entity IDs, relationship types) — strict validation makes sense. But the object field is often descriptive text, not a path component or entity name.
Suggested fix
Use (with a max_length cap) for the field instead of :
This preserves security (null byte and length checks) while allowing natural text in KG object values. The underlying SQLite storage uses parameterized queries so there is no injection risk.
Additional suggestion: bulk KG import
When seeding a knowledge graph from scratch (e.g. mapping a family structure and project states), adding 60+ triples one-at-a-time is slow. A tool that accepts an array of triples would significantly improve the onboarding experience.
Discovered while seeding a KG with family/project data as a daily-use archivist. Happy to PR either fix if welcome. Great tool — using it daily with autonomous cron sessions.