Category: Building Jarvis

Open research notes on building Jarvis — a persistent, self-improving AI agent. One paper at a time.

VRAM Swap: The SSD Trick That Puts a Giant AI on Your Desk, and Who Will Benefit

Can the best open AI models run on the computer you already own? What would it take for you to run the best models on site…
How to write with AI and still sound like yourself

The AI writes a draft in twenty seconds and the editing eats your afternoon. Three changes fixed that for us: marking up a PDF like paper,…
Why your AI assistant forgets what you told it, and how to fix it

Your AI assistant forgot the rule you gave it an hour ago, or the thing you corrected last week. Why that happens at six different moments…
SEAHORSE: What Actually Makes an Agent Code Well

File leases, a bible that asserts shape not liveness, four premises, and the harness. v1.0.
THALAMUS — Which AI Should I Ask?

THALAMUS v4.1 (30 September 2026) is a design for a router that picks the model, the thinking effort, how much of the conversation it sees and…
STRIATUM: Learned Personality Modulation — Tuning an Agent’s Behaviour from Feedback, Not Static Prompt Text

Your agent’s personality is hand-written prose in a prompt file — and the one mechanism meant to tune it from feedback has injected exactly zero bytes…
BROCA: Giving Agent Recipes a Grammar — A Gradual Type System and Combinator Algebra for Self-Composing Workflows

BROCA gives agent recipes a grammar: a gradual typed value model, compile-time contracts, a sandboxed control grammar, a combinator algebra, a skill standard library, catchable recovery,…
SALIENCE: The Death of Fixed Thresholds, the Pyramid of Significance, and Cheap Traversal as the Basis of Next-Generation Vibe Programming

Relevance must be re-computed from the live situation — not read off a frozen list. SALIENCE argues for three pillars: kill fixed thresholds, order knowledge as…
INSTANT RECALL: Finding What the Agent Already Owns

Your assistant often already owns what a task needs, such as a skill, a machine login or last month’s setup, and still doesn’t reach for it;…
Fractal Reasoning: One Reflex at Every Scale, and a Post-Mortem of the Reflection Lane That Has Not Worked Yet

Fractal Reasoning v8.3 (July 2026) proposes one observe-evaluate-adapt reflex for AI agents at every scale, and reports honestly that the reflection lane built to run it…