Open source · User-owned AI context / 2025–2026
OpenContext
Carry your context between AI tools.
A self-hostable context protocol, documented as PCSL, that lets AI tools request selected personal-context namespaces through a local API.
01Context store
The init command creates a local context.json file and signing secret. Namespaces separate identity, preferences, skills, projects, goals, and other context.
02FastAPI server
The local service exposes discovery, authorization, context retrieval, updates, audit logs, and client revocation through HTTP endpoints.
03Scoped token
A client requests named scopes and receives a short-lived JWT. The documented context endpoint returns the namespaces allowed by that token.
04Tool integration
HTTP clients and SDKs fetch context for use in an AI workflow. The repository also includes MCP and browser-extension integration paths.
05Audit & revoke
The CLI provides commands to inspect access logs and revoke client access. Revocation controls future access; it cannot recall context already copied into another tool.
Problem & context
Moving between AI tools often means repeating the same background, preferences, and project details. OpenContext separates that information from any single chat product. Its public repository describes PCSL, the Personal Context Sovereignty Layer, as a protocol for keeping context under user control and sharing selected namespaces.
Context workflow
The documented CLI initializes a local context file, starts the API server, and supports viewing or updating context. A tool requests scoped access, then fetches context using its JWT. The Python SDK can insert the retrieved context into a prompt; raw HTTP offers an integration path for other clients.
Access & deployment
Namespaces make the sharing boundary explicit: an integration can request preferences and skills without requiring the entire context file. Local hosting is the default documented workflow, with optional Docker and cloud deployment. Once context is supplied to an external AI tool, that tool's handling of the data becomes part of the privacy boundary.
Scope & limitations
The README identifies a browser-extension limitation: injecting raw JSON into textareas may not work with every AI chat interface. This case study describes the documented protocol and interfaces; it does not establish a security audit, universal client compatibility, or tested cloud deployment. Scoped tokens still require a correctly configured authorization boundary.