Plectis
This page

Getting started

Overview

Plectis is a runnable public inspection surface for an AI-native workflow system. You run bounded components locally and read what they produced; every component declares the evidence behind it and a scope limit showing where its evidence stops.

What it is

Plectis is the public executable cross-section of a larger AI-native workflow and research runtime: 88 bounded components across formal proof, agent reliability and safety, research and forecasting, projection-drift control, validators, work landing, and continuity. The rest stays private because parts of it are live and handle private data.

The evidence records, local result records, and source links are the accountability layer: they show what each mechanism ran or replayed, what evidence class supports it, and where its authority stops. It is not a hosted service and makes no external model calls.

Plectis is one of two public demonstrations of that system. The other is a standalone Lean repository built around two open Erdős problems, and 13 papers explain the pair as PDFs.

What you can do here

The seven areas

There are 88 components in total. That's a lot to meet at once, so they're grouped into seven areas; the one that matches why you came is the way in:

Next steps

Quickstart →Clone and run a bounded mechanism route. The 13 papers →The active paper corpus, projected from one registry. What it is, in full →A guided read after the evidence loop is concrete. What the pass proves →Evidence kinds, ranks, real-tool markers, and boundaries. Mechanism atlas →The rendered map and scene packet behind the architecture. Components →Browse the generated component index.
Open on GitHub →