Multi-model consensus engine for high-stakes decisions. One query in, structured verdict out, with deterministic replayable deliberation traces and policy guardrails.
Hybrid retrieval engine for security standards and advisories. Combines FTS5 and semantic search with automated CVE/CWE ingestion, and serves results over MCP for in-conversation agent lookup.
Private-by-default self-hosted infrastructure platform. Tailscale-scoped, zero-public-ingress architecture with operational automation across photos, finance, AI, auth, and media services.
Offline prompt-injection evaluation harness. Runs 22 structured attack cases against local models and scores behavior with 8 semantic detectors, with benchmark/regression tracking built in.