Knowledge Bases / Context

Local-first personal knowledge system: highlight what you read, keep notes, chat over your own pages, graph how they connect.

Chrome extension + local dashboard + optional macOS app, one Python backend on localhost:8765. Data stays in ~/.kb/. RAG via Ollama, or OpenAI if you set a key.

Inspired by Karpathy’s knowledge-base post.

Rebecca Joseph & Ramya Iyer

How it fits together

Chrome extension  ─┐
dashboard /app/   ─┼──►  python3 main.py :8765  ──►  ~/.kb/
Context.app       ─┘

Highlight → notes in the side panel → Copy JSON onto your site if you want. Chat and graph sit next to that. Life/Wiki are archived from nav. Default LLM is Ollama; set OPENAI_API_KEY to use OpenAI.

Visuals (More Features to be Explored!)

Chrome extension

Side panel	Chat

Dashboard

Overview	Graph

macOS app

Home	Chat

Setup

macOS, Python 3.11+, Chrome. Ollama or an OpenAI key.

cp .env.example .env          # fill what you use; file is gitignored
python3 -m pip install -r requirements.txt
python3 -m spacy download en_core_web_sm
python3 main.py               # curl http://127.0.0.1:8765/health

• Ollama: leave OPENAI_API_KEY empty; ollama pull qwen2.5:3b and nomic-embed-text.
• OpenAI: set the key; KB_LLM_PROVIDER=auto picks it up.

node scripts/install-launcher.mjs    # starts backend when the extension opens

Chrome → chrome://extensions → Load unpacked → chrome-extension/.

bash scripts/install_app.sh && open -a Context   # optional macOS app

Vars: .env.example. Walkthrough: getting-started.

Tests

pytest tests/

Two tests talk to a live LLM — they fail if that service isn’t up, the rest of the suite doesn’t:

• test_gmail_sync_ingests_threads — ingest still classifies via Ollama (or OpenAI if a key is in .env)
• tests/test_llm_providers.py — OpenAI provider path needs OPENAI_API_KEY (tests stub a dummy key; a real call needs network)

Docs

docs/ · architecture · status · API

Agents: AGENTS.md → init.md

License

MIT © 2026 Ramya Iyer