Hi HN! I built TrustMesh, an open-source reputation system for AI agents.
The Problem: Google's A2A protocol enables agents to communicate across platforms, but there's no standard way to evaluate trustworthiness. When your agent delegates work to another agent, how do you know it won't fail?
The Solution: TrustMesh provides portable trust scores using Bayesian statistics. New agents start at 0.5, scores update with each interaction, recent behavior weighted higher.
Built with FastAPI + SQLite. The trust engine uses Beta-Binomial Bayesian modeling with exponential time decay.
v0.1 alpha shipped today with REST API, Python SDK, and full docs.
Next: Web dashboard, A2A middleware, PostgreSQL support.
Hi HN! I built TrustMesh, an open-source reputation system for AI agents.
The Problem: Google's A2A protocol enables agents to communicate across platforms, but there's no standard way to evaluate trustworthiness. When your agent delegates work to another agent, how do you know it won't fail?
The Solution: TrustMesh provides portable trust scores using Bayesian statistics. New agents start at 0.5, scores update with each interaction, recent behavior weighted higher.
Built with FastAPI + SQLite. The trust engine uses Beta-Binomial Bayesian modeling with exponential time decay.
v0.1 alpha shipped today with REST API, Python SDK, and full docs.
Next: Web dashboard, A2A middleware, PostgreSQL support.
Happy to answer questions!