Say WoT? – Ep. 2: AI, Reputation, and the Death of “Trust Me Bro” with Gzuuus

In Episode 2 of Say WoT?, Avi sits down with Gzuuus to “debug the intersection of AI and the Web of Trust,” tracing his build path through DVMCP, ContextVM, and his latest project, Relatr, with the goal of engineering machine intelligence that amplifies human reputation instead of harvesting it.

Gzuuus explains why ContextVM evolved from an earlier “kitchen sink” approach into a cleaner design that lives at the transport layer, using Nostr keys for authentication/authorization, and leaning into ephemeral events plus a single kind for communications to keep things minimal and composable.

Then the conversation drops into Relatr: a fully open-source, self-hostable Web-of-Trust service designed to run “anywhere” (even on a Raspberry Pi), with real-world performance on a cheap VPS and fast search response times. Relatr supports two complementary interfaces: ContextVM’s request/response pattern (needed for things like search) and NIP-85 Trusted Assertions (a more native produce/consume model that’s easier for clients to integrate).

On scoring, Gzuuus describes a tunable weighted formula that outputs a float from 0 to 1, anchored by social graph distance (hops) mapped through a decay function, with weights configurable by anyone running an instance. Finally, they explore practical client UX (like Amethyst consuming trusted assertions for spam mitigation) and the broader interoperability headaches haunting secure messaging, including MLS complexity and the risk of siloed implementations.

Links

• ContextVM
• Relatr
• Gzuus Github | ContextVM Github
• Enemies of Nostr Article