Plebchain Radio
Plebchain Radio
Avi Burra and QW
Say WoT? – Ep. 2: AI, Reputation, and the Death of “Trust Me Bro” with Gzuuus
1 hour 38 minutes Posted Jan 29, 2026 at 8:46 pm.
Welcome to Say What? With Gzuuus
Gzuuus's first project on Nostr
Colorillo: Contact list app for Nostr
Sovereign Engineering and contributing to Marmot
What exactly is ContextVM?
MCP: A protocol to connect things
Relatr: An open source web of trust solution
Relatr algorithm: A weight formula
DuckDB for Relatr
Publishing scores via trusted assertions
Waterlilly: relay moderated by social dynamics
Explaining rationaling fear with Web of Trust scores
Why did Web of Trust realization take so long?
Plugin system for signals for Relatr computations
0:00
1:38:12
Download MP3
Show notes

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