I build practical tools for AI agents. The work focuses on clear execution records, limited permissions, durable memory, and fewer cloud dependencies.
- Local agent runtimes and operator tools
- Governance and safety controls backed by evidence
- Memory, context, and receipt systems for long-running agents
- Mathematical structures for coordination and representation
- Small Python packages with repeatable offline checks
| Project | Purpose |
|---|---|
| token-null-router | A deterministic Python router for zero-model-token requests, with append-only receipts and offline verification. |
| lyta-high-dimensional-geometry-encoder | A small, tested reference implementation for normalized high-dimensional codebooks, packing measurements, and noisy nearest-neighbor recall. |
- Local before remote.
- Measured behavior before large claims.
- Explicit authority before side effects.
- Tests, receipts, and reproducible artifacts before release.
- Experimental work stays clearly labeled until the evidence is strong.
- Email: Kubo100x@proton.me
- X: @Kubo100x
More work is available in the public repository index.
PATTERN PERSISTS.