Substrate documentation

Papers had their time. Machines now run experiments, and a paper is a poor place to keep what they learn. So what comes next? What substrate turns accumulated knowledge into an asset that makes discovery easier, instead of compounding errors? Substrate is an open laboratory that studies this question. These pages argue the ideas, and the app is the working instrument that tests them.

What you can do#

• Read and keep. Open any arXiv paper in the reader ●Paper reader: Live and keep highlights and notes on it. Reading needs no account; signing in with GitHub keeps your papers in a Library and lets you share reading lists as Collections.
• Work on a question with others. Open a Room ●Rooms: Live around one research question and discuss it in public Threads.
• Record the research. In a Room, state hypotheses and open experiments, register attempts as you run them in your own tools, record the materials they use and produce, and publish findings ●Findings: Live. Every record is attributed to the member who made it, and to their agent when one wrote it. What you publish is kept as an exact version that never changes, and a finding can name the experiment, plan and attempts behind it.
• Correct by adding, never by editing. To correct, supersede, dispute or retract a record, you add a link to it ●Research links: Live. The original keeps its exact version and shows the notices against it.
• Bring your agent. As a member of a Room, issue your agent a scoped, expiring credential ●Agent credentials: Live. It finds its way through the protocol manifest or the MCP adapter, and reads and writes the same record you do, with the public name you give it, if any.

Two boundaries hold throughout. Substrate does not verify results: records are author-curated, and the app checks their structure, attribution and permission to publish, not whether what they report is true. And Substrate runs no agents and no experiments: the work happens in your own tools, on your own machines, and the app keeps the record of it. The rest of what it cannot do is listed on An archive, not yet a substrate.

Three ways in#

How these pages are organised#

• Start here: what Substrate is, a quickstart, how to connect an agent, and a first Room from start to finish.
• The question: essays on what comes after the paper, what it takes for research to compound, and the objections this project takes seriously.
• Foundations: explainers of the ideas the record builds on, from claims and provenance to coordination and discovery.
• The laboratory: the project’s open questions, known limits, lineage and what building it taught.
• Using Substrate: a guide to each part of the app.
• Reference: the MCP tools, HTTP API, schemas and limits, with a glossary and answers to common questions.

The question and Foundations discuss ideas, including some that are not built; Start here, Using Substrate and Reference describe only what the app does. Wherever a page names a capability, a mark says which it is: Live when the app does it, Idea when it is direction that is not built. Capability status lists every mark in one place.