# AutoScholar.ai — Autonomous Research Foundry

From research instructions to verifiable working papers. Research computed on MacLustr.

- `apps/web` — public research library, private admin, HTTP APIs (Next.js)
- `services/scheduler` — durable queue, leases, stage state machine, quality gates, publication
- `services/worker` — MacLustr worker daemon (runs as a dedicated low-privilege user)
- `services/research-runner` — Claude Agent SDK research executor (sandboxed, budgeted)
- `services/qa` — deterministic quality gates · `services/publisher` — transactional publication
- `packages/*` — db (Drizzle/PostgreSQL), schemas (zod), auth, storage, research-contract, maclustr-adapter, observability, ui
- `research/templates` — starter research `CLAUDE.md` files and the LaTeX class
- `docs/` — INFRA_DISCOVERY, DECISIONS, IMPLEMENTATION_STATUS, TEST_REPORT, SECURITY, OPERATIONS, RESEARCH_INSTRUCTIONS_GUIDE, LAUNCH_REPORT

Local development: `cp .env.example .env`, set `DATABASE_URL`, `SESSION_SECRET`, `ADMIN_BOOTSTRAP_PASSWORD`; then `pnpm install && pnpm db:migrate && pnpm dev` (web on :8470) and `pnpm dev:scheduler`.