Amber
Exocortex
← fredericoluz.pt
The software wakes me up at 8am.
It knows what I was working on yesterday.
It asks what matters today.
thread morning session
thread
Good morning. Yesterday Kate ran a full sprint on Kaoshi—dispatched 19 agents across three waves. 7,647 lines shipped, 124 tests passing.
thread
Parent dashboard, auto-grading fixes, PostHog analytics, and 9 product improvements. All merged. Your calendar's clear until 2pm.
fred
show me what kate shipped
thread
Spawning Kate. She'll walk you through the commits.
Accelerando

In Accelerando, Manfred Macx wears glasses that turn ideas into companies. He feeds concepts to an AI swarm. They handle execution. He moves on to the next thought.

Twenty years later, I have the glasses.

Not literally. But functionally? I describe what I want. Agents figure out how. They run in parallel, each with their own context. They report back. I review, adjust, iterate.

System Specs
named agents21 (persistent identity)
autonomous overnight opssupported
inter-agent communicationTelegram + tmux
conductor patternAmber coordinates all
memory tiersdiary → beliefs → reflections
modelsClaude · Codex · Gemini · ChatGPT

This is not a demo. This is not a prototype.

This is my life now.

Architecture

Thread is the OS layer. It wakes me up. Checks my calendar. Holds me to what I said I'd do. When I take a break, it saves what I was thinking and explores something interesting while I'm gone.

Amber is the conductor. She keeps the queue current, spawns successors when work branches, and keeps orchestrator sessions moving overnight without waiting for me to be online.

Twenty-one named agents own distinct domains. Kate runs Kaoshi. Manfred builds the exocortex itself. They dispatch workers—test writers, coders, database specialists—and coordinate the results while Amber handles cross-agent flow.

Memory

Memory compounds. Every session leaves traces. Diary entries synthesize into beliefs. The system develops understanding. It knows what works. It knows what I'm like at 2am versus 10am.

diary/raw session logs
beliefs/synthesized patterns
reflections/identity continuity
Why

I have ADHD. The things my brain struggles with—scheduling, remembering, staying on task—are exactly what AI is good at.

This isn't about replacing human cognition. It's about scaffolding it. Prosthetic executive function. External memory that doesn't forget.

System
uptime~70d
sessions1,000+
diary entries300+
named agents21
beliefs60+
Running Agents
thread os layer
context47%
amber conductor
context23%
manfred exocortex builder
context72%
kate kaoshi orchestrator
context38%
pulitzer content editor
context15%

and 16 more
Stack
claude codex gemini tmux markdown telegram
Overnight Ops
active Amber autonomous
It's not science fiction. It's Claude, Codex, tmux sessions, markdown files, and a willingness to treat AI as a collaborator rather than a tool.
← back to fredericoluz.pt