MediCode v0.1.0 (pre-release)

First public release of MediCode: a zero-cost, split-brain clinical decision-support assistant built on DeepSeek Harness (DSH) (forked at d743267, v0.2.1-alpha.2).

⚠️ Not a medical device. It supports, and does not replace, a licensed clinician's judgement. Patient text sent to the models is processed on Kaggle (Google) infrastructure.

What's inside

• Deterministic clinical math (clinical_calc, 16 calculators: CrCl, eGFR, pediatric Schwartz, BSA, mg/kg dosing, drip rates…). Requests are intercepted before the LLM, and the reply's numbers are verified afterwards.
• Remote GPU brain on Kaggle 2× T4, started from chat with /cloud:
  • II-Medical-8B (GPTQ W4A16 on vLLM, streamed <think>);
  • MedGemma 1.5 4B vision (llama.cpp);
  • Cloudflare tunnel, bearer-token gateway, 3-minute heartbeat, idle self-shutdown;
  • optional Groq-planned boot fixes (never sees clinical content).
• Quality gates /spec /plan /build /test /review /constraints, with a pause-for-confirmation turn gate.
• Context packing with identifier redaction for long pasted records.
• Editorial clinical diagrams (HTML+SVG, 4px grid, WCAG AA, no Mermaid).
• Read-only evidence: PubMed, WHO, CDC and lecture transcripts.

Downloads

File	Use
Source code (zip / tar.gz) below	The complete MediCode harness. Unpack it, then pnpm install && pnpm run build && node apps/cli/lib/bin.js medical.
medicode-overlay-v0.1.0.zip	Only the medical packages plus scripts/apply-to-fork.sh, for applying MediCode to a newer upstream DSH checkout.
SHA256SUMS.txt	Checksums.

Requirements: Node ^22.19 or >=24, pnpm 11, and about 8 GB of RAM for the build. For the GPU side you need a free, phone-verified Kaggle account. Then run /cloud setup, /cloud prepare once, and /cloud start.

Status (why this is a pre-release)

• ✅ 145/145 medical and bundle tests, plus strict TypeScript (noUncheckedIndexedAccess, exactOptionalPropertyTypes).
• ✅ Tunnel auth tests, including an end-to-end run of the real router.py.
• ✅ Applying to a fresh upstream clone passes 5755 tests.
• ⏳ No prebuilt desktop installer yet. The DSH desktop app installs its runtime from npm, so the medical packages must first be published under a MediCode-owned npm scope.
• ⏳ The full monorepo build, a live boot and a real Kaggle GPU session are not yet verified.

See the README and SPEC.md for details and limitations.