openTrinity/mycontext

MyContext – a personal work‑context layer for AI

What it is – MyContext is an open‑source desktop application that continuously gathers a user’s work data (instant‑messaging chats, documents, meeting notes, calendars, approvals, local files, etc.) and builds a private, evolving knowledge graph of the person’s projects, contacts, topics and events. The graph lives on the user’s machine and can be queried by large language models, local agents, or other authorized AI tools so they start with a rich, personal context instead of a blank prompt.

Core capabilities

• Personal context graph – connects people, projects, topics, conversations and facts; lets the user explore relationships and jump back to the original source material.
• Digital self – a “persona” that can read incoming messages, recall relevant background from the graph, and draft replies in the user’s style, but any sending or destructive action requires explicit user approval.
• Search & Q&A – natural‑language queries over the whole work history using a combination of full‑text, semantic similarity and graph queries; answers are assembled by an agent and always include traceable source references.

How it works

1. Channels – unified, permission‑controlled connectors for each work system.
2. Ingest – incremental, checkpointed synchronization that keeps local copies up‑to‑date without re‑reading everything.
3. Store – a local database (user‑scoped) holding raw records, indexes and derived context.
4. Context pipeline – transforms raw records into structured facts, preferences and relationship data.
5. Retrieval – mixes keyword search, semantic recall and graph traversal, always returning source citations.
6. Knowledge graph – stores entities, facts and their relationships for fast queries.
7. Persona – the digital‑self runtime that handles conversation policy, context recall and safe drafting.
8. Agent runtime – isolated sessions that give AI models controlled access to the graph and retrieval services.
9. Desktop – an Electron + React UI that ties all layers together for the end‑user.

Design principles

• Personal context first: the graph is organized around the individual, not the originating app.
• Local‑by‑default: all data and indexes stay on the user’s device.
• Evidence before answers: every AI‑generated answer includes the underlying source references.
• Human‑in‑the‑loop: any irreversible or outward‑facing action (e.g., sending a message) must be explicitly approved.

Current status – The project is under active development. The desktop UI, local storage, search/Q&A, graph, and digital‑self workflow are functional. Support for additional data sources and third‑party integrations is still being added. It is described as a working prototype rather than a finished product.

License – Elastic License 2.0 (source‑available; you may use, modify and self‑host, but you may not offer the software as a hosted/managed service to third parties).