Systems built on Melchizedek

Lyceum Agents is a free interactive course on building reliable agents, available at lyceumagents.com. The curriculum includes technical essays, a glossary, spaced review, and downloadable prompt kits. Every module pairs directly with a reference syndicate file shipped in the framework starter pack.

The public package provides the runnable codebase for the course, and the curriculum provides the operational manual for the package.

melch-research acts as a daily arbiter over news reporting on X. Its ingestion pipeline groups posts by political perspective, validates individual claims against search grounding, and measures divergence between competing accounts. The system publishes a daily edition and an accompanying fact ledger. Public launch is pending.

The pipeline executes Melchizedek syndicates directly. Its Augustin research desk established the verification pattern included in the starter pack.

Operating within a private community, two agent desks manage distinct analytical duties. A financial desk reviews fundamentals, macroeconomic data, and technical charts to publish one morning advisory directive. Alongside it, Augustin acts as a world-events arbiter, answering direct mentions using only its internal research records.

Each bot forwards messages to the A2A server and posts the reply. On the trading side, the governing rule ensures that the model never touches an order.

Cajal is a brain-simulation studio currently in development. The Cajal agent answers questions about neural dynamics by running a whole-brain neural-mass model, a spiking microcircuit, or a pre-trained neuroscience model. The page draws the simulation run beside the text, displaying 85 atlas regions lit on a rotating brain.

The project runs on a syndicate file, the shared model registry, and an MCP server with fifteen tools written against the framework’s tool contract.

Ymir provides an operator console over the framework ledger. It tracks turns, spans, and sampled payloads recorded by active syndicates, tracks trends for individual agents, and dispatches evaluation requests to a judge syndicate to grade specific turns on demand.

Ymir reads the Postgres tables the framework writes and calls back into a syndicate to evaluate an individual turn on demand.