This project renders your provided map and collision-map images as a playable 2D scene.

• map.html (PNG data): main map image
• collision.html (PNG data): collision mask image

From this folder:

python3 -m http.server 8000

Open:

• http://localhost:8000

The game stays client-side, and a local Python server handles Backboard calls so your API key is never exposed in browser code.

1. Install dependencies:

python3 -m pip install -r requirements.txt

2. Set environment variables (do not commit your real key):

export BACKBOARD_API_KEY="<your_backboard_api_key>"
export BRAIN_STATE_PATH="./brain_state.json"

3. Start the brain server:

uvicorn brain_server:app --host 127.0.0.1 --port 8100 --reload

• 1 assistant per in-game NPC personality (obama, spongebob, musician)
• 1 thread per NPC per save file
• Stored in brain_state.json as assistant/thread IDs

• POST /init creates/loads assistants and save threads
• POST /step runs Observe → Decide → Act (tool calls) → Remember

/step accepts structured observation payloads and returns deterministic tool actions (move_to, say, sing, set_goal, set_mood) plus optional narration.

• Move: WASD or arrow keys
• Toggle collision debug overlay: C

• The collision image is rescaled to match the map image dimensions, then converted into a per-pixel collision mask.
• Non-transparent dark pixels in collision.html are treated as blocked.
• Includes 3 wandering NPC agents with random placeholder sprites.
• NPCs use the same collision rules as the player, so they avoid blocked tiles and other actors.
• brain_server.py keeps your Backboard key on the server side and should be run separately from the static map server.