# Paged Context Protocol

An open protocol and Rust implementation for storing, organizing, and retrieving durable context across applications, sessions, and models.

## Metadata

- HTML: https://glenzli.com/en/projects/paged-context-protocol/
- Markdown: https://glenzli.com/en/projects/paged-context-protocol.md
- Collection: Projects
- Language: en
- Published: 2026-08-14
- Updated: 2026-08-27
- Status: active
- Tags: llm-context, protocol, pages, revisions, provenance, memory

## Content

Paged Context Protocol, or PCP, stores, organizes, and retrieves durable context across applications, sessions, and models. It represents content as stable Pages, immutable Revisions, Scopes, Relations, and Provenance, then uses bounded queries to bring selected material into the current task.

PCP can serve one application or multiple tenants contributing to different Scopes under one Identity. The product above it still decides what to record, how to present it, and which results enter the working context.

  ![Durable content moves through identity, revision, provenance, and bounded retrieval into a task](/images/projects/paged-context-protocol-banner.webp)

## Protocol objects

    **Pages and Revisions**
    A Page is a durable addressable record; a Revision is an immutable content snapshot. Source records are usually sealed, while maintained records may be revisioned.

    **Identity, Tenant, and Scope**
    One Store or Runtime serves one durable Identity. Tenants access only authorized Scopes, and the server injects request identity.

    **Relations and Provenance**
    Relations connect stable Pages while evidence and derivation point to exact Revisions. Similarity or temporal proximity does not create a domain relation by itself.

    **Search, Read, and Projection**
    Search returns candidates; callers then read the Payload, Summary, Sources, Relations, or History they need. The Host chooses what to use.

## Current implementation

The repository maintains both the protocol and its Rust implementation: a SQLite Store, embedded and remote clients, Unix socket RPC, Runtime, CLI, MCP server, Console, maintenance coordinator, and read-only facility observation.

  ![PCP Console showing a local Store overview with synthetic data](/images/projects/paged-context-protocol-console.webp)
  The screenshots contain synthetic data and no real Pages, Scopes, or client identities.

The current implementation includes exact, text, graph, temporal, and automatic search; Runtime semantic search and intent matching; Summary, Topic, Validity, Relation, Provenance, archive and restore, lossless packing, and retention planning. Ordinary tenants use `PcpTenantApi`; Runtime maintenance and local administration use the privileged `PcpApi` superset.

  ![PCP Console listing Pages and source information](/images/projects/paged-context-protocol-pages.webp)

## Status and boundaries

The current release is `v0.8.0-draft`. Protocols, interfaces, and Store formats may still change. A v0.7 Store cannot be opened as v0.8; migration requires a new Store and re-import from source material retained by each tenant.

External chats, media, and domain objects remain with the host application; PCP stores their durable context and source references. Semantic search requires an explicitly configured Provider and reports unavailable when none is configured.