← back  the engine

Inside the machine that builds the work.

No mystery box. AJ AI is one machine running one loop — classify → route → execute → persist — over real tools, with a deterministic router and an audit that ends with a human. Here's every moving part.

0 tests passing 0 vault notes 5 execution backends 4 audit gates ~$266/mo to run

01 architecture

One loop, one machine.

Think of it like a tiny operating system. A kernel decides; memory remembers; hands do the work; a clock runs it on a schedule. Every request walks the same four steps.

The kernel is Claude — it reads the request, classifies it, and picks the route. The vault (Obsidian markdown) is canonical memory; SQLite is an append-only ledger for audit, never the source of truth. The hands are Claude Code + Codex, which write and run real code. A scheduler (Windows/NSSM) runs the nightly jobs.

classify → route → execute → persist — every request, same path.
Markdown vault = source of truth · SQLite = audit only.
Runs on one operator, one machine — not a cloud of agents.
Claude kernel classify route execute persist Obsidian vault SQLite ledger Code · Codex scheduler

the kernel decides · memory remembers · hands build · the clock runs it

02 models & routing

One deterministic router. Five backends. Zero added cost.

The router isn't another AI deciding — it's plain keyword/prefix matching that sends each request to the right model. No extra LLM call, no extra latency, and the cheap model handles the cheap work.

Heavy reasoning goes to the big models; extraction and classification go to Haiku; offline/free work can run on a local model. Swapping routine classification onto Haiku cut the monthly model bill by roughly half with no quality loss on those tasks.

Opus · heavy reasoning Sonnet · synthesis Haiku · extract / classify GPT-4o · fallback local · free / offline
router deterministic · $0 Opus Sonnet Haiku GPT-4o local model

keyword/prefix match → exactly one backend, every time

03 the audit moat

Nothing irreversible ships without passing four gates.

Code, deploys, anything touching money — every one runs the gauntlet. Three machine checks, then a human. That's why you can trust what comes out.

01Deterministic lintssyntax + safety rules, no model needed
02LLM reviewClaude checks logic + intent
03Independent audita different model, read-only, second opinion
04Human sign-offAJ approves — or it does not ship

The moat earns its keep: on the betting pipeline it caught a bug mis-grading 68% of picks as fake losses, then applied 496 production corrections — before any of it counted.

04 the stack

What actually runs it.

decideClaude — the kernelclassify + route
memoryObsidian vault + SQLite ledgertruth + audit
handsClaude Code + Codexwrite + run real code
connectMCP toolsGitHub · Linear · Playwright · Context7
clockWindows scheduler (NSSM)nightly jobs

honest by design

Built to be trusted because it audits itself.

One operator, one machine, a few hundred dollars a month — and a paper trail for every irreversible thing it does.