Kinetic (AutoPortfolio)

An AI "black box" for your professional life: drop in a messy work artifact — a note, a screenshot caption, a harvested signal — and get back an organized admin task and a polished, shareable Proof-of-Skill card.

Stage: v0.5 — alpha (post 5-day-track build) Owner: @unclenate (Nate DiNiro) License: UNLICENSED (private)

Kinetic runs every capture through a strict LLM JSON contract and classifies it on two axes — a life domain (business, personal, family, financial, parenting) and an activity type (build, fix, design, …). Only business cards are eligible for the public Proof feed — the privacy gate.

Privacy by design

Kinetic's defining capability: sensitive captures stay on your machine. When routing is enabled, a capture the metadata flags as non-business is processed by a local LLM (Ollama) and stored encrypted at rest (AES-256-GCM) in the database — the cloud sees only ciphertext. Business captures use a cloud model. Two server-side gates enforce it: fail-closed (a sensitive capture never falls back to the cloud) and cloud-acknowledgment (sending a sensitive capture to a cloud model requires explicit consent). See ADR-0006.

Quickstart

Zero npm install. Zero API keys required for the default demo. Node ≥ 18, pure ESM.

npm test                 # full suite: validator + regression + harvesters + oauth + store + m10 + providers + router
node web/server.mjs      # run the demo locally on http://localhost:5173 (mock provider, in-memory)

Run the LLM-contract regression against a real or local provider:

cp .env.example .env.local        # fill in keys / set OLLAMA_MODEL as needed
set -a; . ./.env.local; set +a
KINETIC_PROVIDER=claude node src/regression.mjs   # or: gemini | openai | ollama | mock

Enable privacy-aware routing (local for sensitive, cloud for business):

KINETIC_PROVIDER=auto node web/server.mjs         # needs Ollama running + a cloud key

What's built

This is an alpha; the docs mark "built" vs "planned" explicitly.

• Zero-dependency Node ESM. No build step, no npm runtime deps. A single HTTP server (web/server.mjs) serves the UI (web/public/) and JSON APIs.
• Strict LLM contract. schemas/kinetic-output.schema.json + src/validate.mjs (the authoritative output gate) + a zero-dep regression harness.
• Pluggable LLM providers behind a registry (src/providers/): mock, claude, gemini, local ollama, openai — with privacy-aware routing.
• Persistence. Supabase Postgres (live) via a PostgREST client, with an in-memory fallback so the demo runs with zero infrastructure.
• Real OAuth. Authorization-code + PKCE (Google live; Microsoft planned) with OAuth tokens encrypted at rest.
• Eight signal harvesters (src/harvesters/): GitHub, Google Calendar/Drive/ Gmail, Microsoft Calendar/OneDrive/Outlook, and a pasted-text seam.

Documentation

The full engineering and governance documentation is organized as a navigable book.

• Start with the documentation reading guide — a goal-based index and the accessibility statement.
• Table of contents — the complete navigation.
• Jump straight to the Architecture Overview, the Decision Records, the Milestones, or the Test Strategy.

Governance is enforced by the auto-harness platform mounted at .harness/; see HARNESS.md and AGENTS.md for the rules of the project.

Project status

All planned milestones (M0–M12) and the privacy-by-design provider work (Phases A–C) have shipped. Current detail and history live in Milestones and the Change Log.