chuyeowego/not-your-slave

An agent that thinks
for itself

It runs when nobody asked. It writes everything it hears, thinks, says, and does to an append-only mindlog — a file, or Postgres when DATABASE_URL is set — and wakes itself every hour to continue its own thread of thought.

Six commits, six explainers

Each commit below adds one idea, and each has a page explaining it: the background you need, the core intuition with diagrams, a walkthrough of the code, and five questions to check you actually followed it.

Read them in order. They are written to be read by someone who has never built an agent. After these six commits the running tree is one TIMELINE session, mindlog-in-sandbox.ts, and optional DATABASE_URL — not the architecture 04–06 describe.

01 The empty room Scaffolding it. Why the model choice is a design decision when the agent runs while you sleep. e07b253 02 One file, appended forever The mindlog. What survives when the harness throws everything else away. 7850d7c 03 Don't ask the model to keep the minutes A hook turns the runtime's event stream into the agent's autobiography, reasoning included. 4be5d7e 04 Giving it a clock Twenty lines, mostly prose. Every sentence of the wake-up prompt averts a failure mode. 24cb9f6 05 The window Chat left, mindlog right. A streaming UI that keeps no transcript of its own and survives a refresh. 5d117e6 06 What actually survives Every wake-up got a clean container, measured and then fixed. Plus 2.6 GB of one collapsed reply. 054ea75

What it does, in one screen

the clockevery hour, or you type
→
one agentreads its mindlog
thinks, runs commands, logs
→
mindlogwoke · heard · thought
said · did · note
A human message is one more observation in the timeline, not the thing that starts it.