PALEO · PoT companion (local)
Runs beside Path of Titans — no game files changed Site home Profiles Technical Agent Visual Agent Architecture Gallery Agent demo
Recommended runtime: PALEOOverlay.exe is the main in-game runtime (always-on-top capture + debug beside Path of Titans). PALEO.exe runs this Companion HUD in your browser (local Python server + Instinct Agent) for monitoring and manual testing. Use sliders or live capture when enabled, then Run Instinct tick / Auto. Full keyboard/mouse control is optional via run_paleo_control_loop.py (see README) with F12 emergency stop.
Run the real Instinct Agent on your PC

A button on this website cannot start Python on your computer (every browser blocks that for safety). For outputs that match simulate_instinct_decision in the repo, run the small local server below once per session.

  1. Recommended in-game runtime (overlay):
    py -3 scripts/run_paleo_overlay.py --mode advice --classifier-checkpoint results/experiments/serengeti_disk_resnet18/resnet18_serengeti_disk.pt
    For guarded keyboard control tests, use py -3 scripts/run_paleo_overlay.py --mode control --enable-control --classifier-checkpoint results/experiments/serengeti_disk_resnet18/resnet18_serengeti_disk.pt (keep F12 emergency stop ready).
  2. PALEO.exe-style HUD runtime (browser companion):
    py -3 scripts/run_paleo_live.py --classifier-checkpoint results/experiments/serengeti_disk_resnet18/resnet18_serengeti_disk.pt
    This starts the local Companion HUD server and opens this page with live capture + classifier path.
  3. Open a terminal (PowerShell, Command Prompt, Terminal, etc.) and cd to your PALEO repository root — the folder that contains scripts/ next to src/ and pages/.
  4. Start the companion server. In that same terminal, run exactly: python scripts/serve_companion.py (default port 8765). Leave this window open while you use the HUD.
    python scripts/serve_companion.py
    Optional classifier wiring for live HUD threat: run python scripts/serve_companion.py --classifier-checkpoint results/experiments/serengeti_disk_resnet18/resnet18_serengeti_disk.pt (path should point to your trained checkpoint). Port already in use? Use python scripts/serve_companion.py --port 8766 — then in step 3 replace 8765 with 8766 everywhere (or use Copy URL after you opened this page on that port once).
  5. In your web browser, go to the local address below — this is on your machine (127.0.0.1), not your GitHub Pages URL. Type or paste: http://127.0.0.1:8765/companion-hud.html
    http://127.0.0.1:8765/companion-hud.html
  6. You should see this same Companion HUD, with no yellow “static mode” banner. Click Run Instinct tick; the gray status line should end with (live API). If the banner is still there, the tab is not talking to your local server — double-check step 2 is running and step 3’s URL matches your port.

Telemetry

Threat (predator probability) 0.35
Prey density 0.40
Health 0.85
Stamina 0.75
Hunger 0.45
Thirst 0.35

Full fidelity: use the “Run the real Instinct Agent” steps above (copy buttons + terminal). This panel is the offline preview when the API is missing.

Thought log

Ready.
{ }
Live captured frame preview
debug: waiting...

Primal Mind · facts

Diet

Size / role

Environment

Notes

Instinct Agent chooses a mode from telemetry; Primal Mind is the structured identity/memory model in-repo (docs/lexicon.md).