MARY KIN BETA — FRESH WINDOWS LAPTOP INSTALLATION CANDIDATE
Marykin Labs • Local-first companion • Private beta • 4 November 2026

THIS IS NOT THE MARKETING WEBSITE: the installed Mary app is the private local chat + Work/Results interface.
It is a separate, frozen Windows application; it does not need the PythiaGrid source, Python or Node on your laptop.

BEFORE INSTALLING
- Windows 10 22H2 or newer, 64-bit; enough disk for Ollama and selected models.
- Install and start Ollama from its official website, if it is not already present:
  https://ollama.com/download/windows
- Verify in PowerShell: ollama list
- If Ollama is already installed, reuse your models; do not delete or download them again.
- Do not disable Windows Defender, firewall, SmartScreen, BitLocker or other protection to make this beta run.
- The ZIP includes Mary executable and UI, but it does NOT bundle Ollama or any model weights.

FRESH INSTALL
1. Close any old Mary version currently using 127.0.0.1:8128. Do not uninstall it or remove existing data.
2. Download the qualified Mary-Kin-Laptop ZIP from the authenticated private release or the verified ChatGPT attachment.
3. Check the ZIP's published SHA-256 against the matching release evidence before extracting it.
4. Extract the ZIP to a normal folder (not inside OneDrive sync, Program Files or a temporary archive viewer).
5. Double-click INSTALL-MARY.cmd. It verifies every bundled file and the frozen source identity.
6. Open the new Mary Models desktop shortcut. Select an already-downloaded model (suggest qwen3.5:4b on the 24-GB laptop if present; otherwise an installed 2–4B model). A download requires explicit YES.
7. Open the new Mary Beta desktop shortcut. It launches http://127.0.0.1:8128/ on this laptop.
8. In Mary, send a short text message, refresh, try a second conversation, then test Work & results with a safe sample file.
9. If local voice appears, speak, stop, inspect/edit transcript, choose Send to Mary or Discard. Voice controls appear only when the existing governed voice bridge is ready.

WHAT YOU SHOULD SEE
- 'Mary Core ready' means local Core HTTP is responding; it does not yet certify Ollama or a model.
- A real model reply in chat confirms the local Ollama → Mary chat path.
- Work & results shows only supported approved capabilities; not arbitrary remote PC control.
- Ordinary chat deletion and Work evidence remain separate.
- Private local data persists under your user profile; no login to the founder workstation is needed.

FAILURE CASES
- If the Mary Models shortcut says Ollama is unavailable: launch Ollama and verify http://127.0.0.1:11434/api/tags in the laptop's browser, then retry.
- If 'another Mary is running': close that version; don't kill or replace a service to force the port.
- If the model response is slow: choose an installed smaller model (2–4B) in Mary Models.
- If the microphone/voice bridge is unavailable: text chat should remain usable; do not substitute unapproved cloud TTS.
- If Windows flags unknown publisher: do not bypass security; the beta needs separate signing/SmartScreen acceptance.
- If anything fails: stop and save the exact onscreen error, app version/receipt and safe redacted logs; avoid repeated reinstallations.

ACCEPTANCE BEFORE LABEL 'READY FOR THIS LAPTOP'
Real laptop: secure extract + digest PASS, app first run PASS, Core READY, Ollama reachable, configured local model answers, 2 chat turns and restart persistence, a permitted sample Work task with independent verdict, no data loss, keyboard/mobile layout checks, and voice transcript-before-send when local bridge exists. Signing/SmartScreen and any requested background-mode acceptance are separate gates.

No claim of completed laptop physical acceptance is made by this file. This candidate must pass fresh Windows CI first and then the real laptop smoke test.
