FORGE STACK — SaaS Forge pod NOW · vault PARKED
Jane gym · privacy lane · not 160GB monolith

## GRAB FORGE (Brian · today)
Rent 24GB pod · stop when idle · keys on pod disk only
→ `FORGE_GRAB.md` · `bash ~/jane/forge_grab.sh` · forge gym **Grab Forge**

**No GitHub vault.** `~/.stan/` on **your** pod. Linode = voice home.

## LINODE (keep)
| pipe | where | $ |
|------|-------|---|
| voice talk | Groq on Linode apex | ~$5–20/mo |
| voice speak | Edge Jenny TTS | ~$0 |
| hands | Aider + OpenRouter on Linode | ~$10–40/mo |
| match | Vosk CPU on Linode | $0 |
| public | Linode + Cloudflare | ~$24–48/mo |
| ship | bash bots tier 0 | $0 |

Brian uses: jane.hitme.dev · forge.hitme.dev · gym taps · no VRAM purchase.

## LATER (when Brian buys box)
LAW: one pipe hot at a time · load → job → unload

| pipe | OSS | VRAM | job |
|------|-----|------|-----|
| ears | whisper.cpp / faster-whisper | ~2-4GB | STT local |
| mouth | Piper TTS | ~0-2GB | Jenny-class voice |
| brain | Ollama 7-8B Q4 (Qwen/Mistral ok) | ~5-8GB | quick talk · private |
| match | Vosk speaker_id | CPU | Brian voice lock |
| hands | Aider + bash bots | 0 | ship · no GPU |

NOT IN SCOPE: 70B local · $1200/mo cloud 160GB · Cursor as spine

INSTALL ORDER (when box lands):
  1. nvidia-driver + nvidia-container-toolkit
  2. ollama · pull qwen2.5:7b-instruct-q4_K_M
  3. faster-whisper medium
  4. piper (en_US-lessac-medium)
  5. forge_scheduler.sh — load/drain wrapper

PUBLIC: https://forge.hitme.dev/
Word: forge · gym · 24gb · oss