Hi HN — I built Mullion, an OSS TS toolkit to prevent LLM context leaks (admin/user, internal/external boundaries).
- Owned<T, Scope> tags data with trust scope
- scope crossing requires explicit bridge(...)
- ESLint rules flag cross-scope use / leaked values
- OpenTelemetry-compatible tracing + cost estimation
integrates with Vercel AI SDK
npm i @mullion/core @mullion/ai-sdk
https://github.com/mullionlabs/mullion-ts
Feedback welcome: does this model match how you handle trust boundaries in production?
