All projects

synod

0 comments

What it does

synod is a concurrent research swarm. You ask one question, and instead of a single model answering, a small team of agents works on it together: they search the web, verify what they find, and write a clean, cited answer, all at the same time. It turns a plain question into a sourced research brief in a few seconds, where every claim links back to a real source. And you don't just get the result; you watch the swarm work live in the console, one board per role.

How the agents run concurrently

synod runs six agents on one Mozaik runtime, coordinating purely through events on a shared bus, with no central scheduler. A Planner splits the question into several angles and emits one search event per angle. Three Legate scouts pick those up and search in parallel; each Legate is model-driven and decides its own queries. An Assessor reacts to every finding the instant it lands and verifies it, without waiting for the search to finish. A Scribe folds verified findings into the answer continuously, and if coverage is thin it emits an event back to the Legates asking for more, so the flow is a loop, not a line. All of them share one typed RuntimeState, and a dedicated Observer participant watches the bus and records a live timeline you can see in the console, where the agents visibly overlap in time. Discovery, verification and synthesis genuinely happen at once.

Comments

No comments yet.

Sign in to like and comment

Newsletter

For developers who want to learn how to build self-organizing agents.

Join our online hackathon