Project Context Protocol

A portable context architecture for long-running work with any LLM.

Version 1.1.0 · Created by Omnipressence

Project Context Protocol (PCP) turns an AI project from a collection of disconnected chats into a structured working environment built around four durable context types:

• Instructions — how work must be done
• Knowledge — facts, references, specifications, architecture
• Skills — repeatable procedures and governed methods
• State — the verified current checkpoint, decisions, constraints, and next action

The protocol is LLM-agnostic. It can be used with ChatGPT Projects, Claude Projects, Gemini, local models, coding agents, or any system that can read project files.

Why it exists

Long-running AI work usually breaks for one reason: the model has access to a lot of context, but no reliable way to know what is authoritative now.

PCP solves that by separating temporary conversation from durable project truth.

Chats are working sessions. Project files are durable operating context.

Core model

PROJECT
├── Instructions
├── Knowledge
├── Skills
└── State
     ├── Current checkpoint
     ├── Decisions
     ├── Constraints
     └── Next action

The model retrieves only the context required for the current task and follows an explicit authority order when sources conflict.

What this improves

• Less repeated prompting between chats
• Faster recovery when starting a new session
• Lower context clutter
• Fewer contradictions with earlier decisions
• Better continuity across models
• Clear separation between facts, instructions, procedures, and transient conversation
• Easier human review of what the AI believes the current project state is
• Better handoffs between people and models
• More predictable use of project-specific skills and constraints

5-minute setup

1. Copy protocol/PROJECT_CONTEXT_PROTOCOL.md into your AI project.
2. Copy templates/CURRENT_STATE.md into the project and fill in the current checkpoint.
3. Add DECISIONS.md and CONSTRAINTS.md when the project needs them.
4. Add your existing reference documents, specifications, and skills.
5. Tell your LLM:

Use PROJECT_CONTEXT_PROTOCOL.md as the governing context protocol for this project.
Before substantive work, recover the relevant current state, decisions, constraints,
knowledge, and skills. Prefer the latest verified project state over older conversation.

Then start new sessions with:

Continue from the current project state.

Recommended project structure

project/
├── PROJECT_CONTEXT_PROTOCOL.md
├── CURRENT_STATE.md
├── DECISIONS.md
├── CONSTRAINTS.md
├── ARCHITECTURE.md
├── SKILLS/
├── SPECS/
└── REFERENCE/

Only create files that materially improve continuity. The protocol does not require every folder.

Use with any LLM

See docs/LLM_SETUP.md.

Human workflow

See docs/QUICKSTART.md.

Design principle

The goal is not maximum context.

The goal is:

Minimum sufficient authoritative context.

That means orient first, retrieve only what is relevant, go deeper only when necessary, and preserve durable state separately from chat history.

Repository contents

• protocol/PROJECT_CONTEXT_PROTOCOL.md — portable governing protocol
• templates/ — ready-to-copy project state files
• docs/ABOUT.md — what the protocol is
• docs/BENEFITS.md — why it improves long-running AI work
• docs/LLM_SETUP.md — instructions for ChatGPT, Claude, Gemini, Codex, local models, and other LLMs
• docs/QUICKSTART.md — minimal operating workflow
• docs/ARCHITECTURE.md — context architecture and authority model
• ui/content.json — structured copy for a future interface or onboarding experience
• assets/logo.svg — official repository logo
• VERSION — current release version
• CHANGELOG.md — release history

Versioning

PCP follows semantic versioning: MAJOR.MINOR.PATCH.

• MAJOR — breaking protocol or file-contract changes
• MINOR — backward-compatible capabilities, templates, or workflow improvements
• PATCH — documentation, wording, and non-breaking corrections

Current release: 1.1.0

See CHANGELOG.md for release history.

Inspiration and attribution

Project Context Protocol was inspired in part by the context-engineering ideas explored by OpenViking, particularly the separation and structured retrieval of memory, knowledge, skills, and working context. PCP is an independent, lightweight, LLM-agnostic project designed for file-based use inside consumer and developer AI workspaces rather than a fork or reimplementation of OpenViking.

Original project: https://github.com/volcengine/OpenViking

Created by Omnipressence

Project Context Protocol is maintained by Omnipressence, focused on governed AI systems, automation, and practical AI operating environments for business.

Website: https://omnipressence.com

Status

Version 1.1.0 establishes the branded, versioned public release structure while preserving the portable context protocol and project templates. It intentionally does not depend on a specific model, API, vector database, or vendor-specific memory system.