A NEW WAY TO BUILD SOFTWARE WITH AGENTS

Software is a living graph. Code is an artifact.

Agents forget everything between sessions. Precursor is the layer that remembers — it interviews you, writes every decision, constraint, and open question into a versioned graph, and your agents build from it. Governed. Traceable. Compounding.

Get early access MACOS · YOUR AGENTS, YOUR SUBSCRIPTION

GOAL DECISION · CONFIRMED CONSTRAINT QUESTION MODULE CODE — COMPILED FROM THE GRAPH

01 — THE PROBLEM

Agents don't lack intelligence. They lack memory.

A human engineer onboards once — every decision, every review, every lesson compounds for years. An agent starts from zero, every session. And every tool we hand it — files, tickets, docs — assumes a builder who remembers. That is why unattended codebases rot.

02 — THE PRIMITIVES, RETHOUGHT

HOW YOU BUILD

You don't prompt. You decide.

Precursor interviews you, the way a great engineer would. Every open decision is answered by you, explicitly delegated to the AI, or routed to whoever knows. Nothing is left ambiguous — ever.

WHAT IT PRODUCES

The graph is the product.

Every decision, constraint, and question — versioned, linked, traceable to who made it and why. Agents pull the exact context each task needs; everything they do writes back. Code is compiled from it.