Skip to content

Project Corpus

Durable project identity, continuity and authority for AI-assisted work.

AI sessions end; the project should not have to start from scratch. Project Corpus is a Markdown-first, vendor-neutral protocol that records what a project is, where it is now, what has been verified, what comes next, and which rules govern change.

Project Corpus connects one AI session to the next through canonical project state.

In 30 seconds

Project Corpus gives a project an owned, inspectable source for identity, current state, evidence, authority, and next actions across AI sessions. The Protocol works as ordinary Markdown with no installation, database, or MCP server. An optional local Runtime adds validation, external owner trust, policy enforcement, verified writes, audit/recovery, CLI, and a local stdio MCP adapter.

Start manually

Copy the V2 minimal template into a project, fill PROJECT.md and STATUS.md, then give the agent the project folder or upload the canonical files. The Quick Start has the exact paths and a ready-to-use loading prompt.

Choose the layer you need

Project Corpus Protocol Optional Project Corpus Runtime
Markdown-first and vendor-neutral Controlled CLI and local stdio MCP
No installation required External owner trust and policy enforcement
Manual and direct-folder workflows Verified transactions, audit, and recovery

The Runtime implements the Protocol; it never silently defines or broadens it. For security and platform boundaries, start with the security model.

Higher-level lifecycle, recovery, or orchestration systems can build on Project Corpus while retaining their own contracts. Living Software Organism is one independent reference direction; Project Corpus is not LSO and does not require it.

Context Intelligence 2.2.0

Canonical Markdown → Rebuildable Context Index → Authority/temporal retrieval → Bounded Context Bundle → Any AI agent.

Markdown remains the truth. Context Intelligence helps an agent find the right truth. This optional Runtime feature returns source-backed, non-authoritative context; the agent must read the primary files before making authority claims. See the Context Intelligence guide.

Existing V1 projects

V1 remains supported. The V2 migration path plans changes read-only and creates a separate V2 destination without rewriting V1 templates. See Migration.

The documentation site is an English-first reference. The repository keeps Russian documentation alongside English source documentation.