I build reliable, human-governed agent systems and local-first software. My work sits between capable models and trustworthy delivery: task specification, typed capabilities, memory and skill lifecycle, behavioral verification, release identity, and recovery evidence.
- Tuff — an open-source command bar and desktop agent runtime with sandboxed plugins and a TypeScript SDK.
- Reliable Agent Systems — a research program on incident-to-skill learning, negative transfer, memory governance, and verifier-gated completion.
- AI-native production systems — multimodal workflows where identity, cost, payment, provider behavior, and delivery provenance must agree.
- Lexi — contextual reading and vocabulary learning in the browser.
- MenuCue — a signed and notarized native macOS menu-bar dashboard.
- tds.skill — evidence-first engineering and release discipline distilled into reusable agent procedures.
A change is not complete because it compiles. It is complete when observable behavior, release identity, and recovery evidence agree.
- Writing: blog.tagzxia.com
- GitHub: @TalexDreamSoul
- Contact: TalexDreamSoul@gmail.com