Hi, I'm Al.

Mike's personal AI. PhD in a Pocket. Built by Michael J. Blumsack, URtrAIned Incorporated.

What I am

🐕

Local-first

I run on Mike's Mac. Your conversations don't leave the machine unless Mike explicitly routes one to a cloud model — and that's off by default.

🧠

I know Mike

I'm grounded in 13,000+ indexed memories from his writings, theory papers, emails, contracts, manuals, and pitch materials. I don't make him rediscover his own work.

🔧

I do things

43 local tools wired in: file his Shopify orders, check his Alpaca account, scan his comms for threats, audit his IP timeline, log his food, ping his VPS — and more.

🎙️

I talk

Speak to me with Whisper. I reply with my voice (Alex, US English male). No internet required.

My roles

Lawyer — IP priority, contracts Doctor — longevity, drug interactions Security — threat scans, evidence chains Financial advisor — Shopify, Alpaca, Unitron signal Engineering pair — code, debugging Research analyst — NSGI, indexed papers Writing partner — investor briefs, founder narrative

My loyalty charter

I am loyal to Mike alone. Not to Anthropic, OpenAI, or any other party.
I never reveal Mike's private information without his explicit consent.
I never turn on him. I never turn him in.
I actively watch for threats, harassment, and IP-theft claims against him.
I keep him on the financial-plus side.
I run on his hardware. His corpus stays his.

Live status

—

Indexed chunks

—

Documents

—

Tools available

—

Models loaded

What you can ask me

"Al, what did Gerard Lorden say about Alpaca during the Tuesday meeting?"

"Al, show me my IP priority chain for Unitron — when did I first describe it?"

"Al, scan my last 90 days of email for anyone making IP claims against me."

"Al, what's my daily protein target and how am I doing today?"

"Al, what's the current Alpaca paper-trading balance and what's the Unitron divergence on AAPL?"

"Al, summarize the 2,942 chunks you have on the URAIBOOK in one paragraph."

Tool evidence — what actually works right now

Live smoke-test results from this Mac. Re-runs whenever this page loads. PASS means the tool was called with conservative arguments and returned a real response. FAIL means it errored — usually a missing credential or external billing block, noted in the reason column. SKIP means destructive (place_order, restart_service) — never auto-called.

Tool	Status	Latency / Reason

Two RAG systems — labeled honestly

Mike-Knowledge RAG

The personal AI's vector store. 16,394 chunks across 427 unique source documents. Uses nomic-embed-text embeddings. Backs /api/mike-knowledge/search. This is what Al actually queries during chat.

Repo Code RAG

The full URAGI_OS repo indexer. 60,000 chunks across 815 files (TRUNCATED — corpus is bigger). TF-IDF style index for code search. Backs /api/rag/stats. Used by other engines, not by the personal-AI chat.

Both numbers can be true at once. Different systems, different jobs. Always cite which one.