Building local-first agent systems, trust-zone memory, and practical anti-extraction infrastructure.
Local-first assistant runtime for bounded memory, trust zones, and accountable continuity.
| Runtime | Safety | Memory | Public Mode |
|---|---|---|---|
/health, /prompt, /memory/graph |
loopback default, auth gate, rate limiting | scoped retrieval | paid/public ephemeral by default |
flowchart LR
User["Operator"] --> Server["Local Runtime"]
Server --> Prompt["Prompt Pack"]
Server --> Memory["Bounded Memory"]
Server --> Tools["Explicit Tools"]
Server --> Zones["Trust Zones"]
Zones --> Private["private_self"]
Zones --> Paid["paid_public ephemeral"]
- Local-first agent runtimes with explicit trust boundaries
- Memory systems that separate private continuity from public/client surfaces
- Operational readiness gates for APIs, hosting, rate limits, caching, scaling, errors, and accessibility
- Governance and mechanism design for anti-extraction systems
- Practical repo discipline: tests, runbooks, status vocabulary, and launch proof
- Making Dizzy visually legible and technically credible in public
- Turning readiness requirements into audited project surfaces
- Building systems that prove what runs today before making larger claims
For client-facing or hosted work, I track:
| Area | Gate |
|---|---|
| Front end | Minified production build, no secrets, source maps intentionally gated |
| Database | RLS or tenant isolation on authenticated reads/writes |
| Version control | Protected main, reviewed deploy commits, tagged releases |
| APIs | Auth, validation, CORS, request limits, structured errors |
| Hosting | HTTPS, least-privilege env vars, health checks, rollback path |
| Rate limiting | Per-user or per-IP limits on public/expensive routes |
| Caching | Static assets cached, sensitive responses protected |
| Scaling | Queue/backpressure and outage behavior documented |
| Error tracking | Scrubbed server/client errors with release context |
| Accessibility | WCAG 2.2 AA target; at least WCAG 2.1 AA where required |
Meme:
Vibe: architect of post-extraction cognitive infrastructure
Evidence: tests, runtime endpoints, readiness gates, and receipts