Add OpenEnv API endpoints and fix reset/step/state support for Phase 1 checks

Dockerfile

@@ -16,7 +16,7 @@ COPY ui/ ./
 # HF Spaces forwards the matching secrets as Docker build-args automatically
 # when you set them in Space Settings → Variables and secrets.
 ARG VITE_HF_TOKEN=""
-ARG VITE_HF_MODEL="mistralai/Mistral-7B-Instruct-v0.2"
+ARG VITE_HF_MODEL="meta-llama/Llama-3.1-8B-Instruct"
 ENV VITE_HF_TOKEN=${VITE_HF_TOKEN}
 ENV VITE_HF_MODEL=${VITE_HF_MODEL}
 

openenv.yaml

@@ -26,11 +26,11 @@ env:
 
   variables:
     - name: HF_MODEL
-      default: "mistralai/Mistral-7B-Instruct-v0.2"
+      default: "meta-llama/Llama-3.1-8B-Instruct"
       description: >
         Any HF chat/instruct model compatible with the [INST] prompt format.
-        Alternatives: HuggingFaceH4/zephyr-7b-beta
-                      meta-llama/Llama-3.1-8B-Instruct (needs license accept)
+        If you prefer a different model, use HuggingFaceH4/blackbird-7b or
+        meta-llama/Llama-3.1-8B-Instruct (the latter may require license acceptance).
     - name: PORT
       default: "7860"
       description: Port exposed by the Docker container (HF Spaces default).

server/app.py

@@ -22,6 +22,7 @@ from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import JSONResponse
 
 from server.environment import GameState, InferResponse, build_prompt
+from inference import SurvivalIslandEnvironment
 import models
 
 # ── Logging ───────────────────────────────────────────────────────────────────
@@ -101,6 +102,9 @@ async def root():
     }
 
 
+env = SurvivalIslandEnvironment()
+
+
 @app.get("/api/config")
 async def config():
     """Public runtime configuration for the frontend."""
@@ -111,11 +115,39 @@ async def config():
     }
 
 
+@app.post("/openenv/reset")
+async def openenv_reset():
+    """Reset the OpenEnv environment and return the initial state."""
+    state = env.reset()
+    return {"state": state}
+
+
+@app.post("/openenv/step")
+async def openenv_step(payload: dict):
+    """Advance the OpenEnv environment with a given action."""
+    action = payload.get("action")
+    if not action or not isinstance(action, str):
+        raise HTTPException(status_code=400, detail="Missing or invalid action")
+    next_state, reward, done, info = env.step(action)
+    return {
+        "state": next_state,
+        "reward": reward,
+        "done": done,
+        "info": info,
+    }
+
+
+@app.get("/openenv/state")
+async def openenv_state():
+    """Return the current OpenEnv environment state."""
+    return {"state": env.state()}
+
+
 @app.post("/api/infer", response_model=InferResponse)
-async def infer(state: GameState, request: Request):
+async def infer(request: Request):
     """
-    Accept a GameState JSON body, build the LLM prompt,
-    run inference, and return the chosen action + thought.
+    Accept either a raw prompt or the frontend GameState JSON body.
+    Build the LLM prompt server-side and return the chosen action + thought.
 
     Inference priority: local pipeline → HF Inference API → rule fallback.
     """
@@ -126,10 +158,17 @@ async def infer(state: GameState, request: Request):
             detail="Too many requests — wait a few seconds.",
         )
 
-    prompt = build_prompt(state)
+    data = await request.json()
+    state = None
+    if isinstance(data, dict) and "prompt" in data:
+        prompt = data["prompt"]
+    else:
+        state = GameState.model_validate(data)
+        prompt = build_prompt(state)
+
     logger.info(
-        f"[infer] gen={state.generation} ip={ip} "
-        f"challenge={state.activeChallenge.type if state.activeChallenge else 'none'}"
+        f"[infer] model={os.getenv('HF_MODEL')} ip={ip} "
+        f"challenge={state.activeChallenge.type if state and state.activeChallenge else 'none'}"
     )
 
     # Try inference paths

ui/.env.example

@@ -25,4 +25,4 @@ VITE_HF_TOKEN=YOUR_HF_TOKEN_HERE
 #
 # Requires license acceptance on HuggingFace.co:
 #   meta-llama/Llama-3.1-8B-Instruct
-VITE_HF_MODEL=mistralai/Mistral-7B-Instruct-v0.2
+VITE_HF_MODEL=meta-llama/Llama-3.1-8B-Instruct

ui/src/App.jsx

@@ -13,10 +13,8 @@ const TICK_RATE = 1000;
 const WORLD_WIDTH = 6000;
 const WORLD_HEIGHT = 3000;
 
-// ─── HuggingFace config from .env ────────────────────────────────────────────
-const HF_TOKEN = import.meta.env.VITE_HF_TOKEN;
-const HF_MODEL = import.meta.env.VITE_HF_MODEL;
-// ─────────────────────────────────────────────────────────────────────────────
+// ─── Runtime backend config ─────────────────────────────────────────────────
+// Token is kept server-side; the frontend calls /api/infer and reads /api/config.
 
 const ZONES = {
   OCEAN: { baseEndX: 400, color1: '#094b65', color2: '#20a4c0' },
@@ -243,11 +241,30 @@ export default function App() {
 
   const [hudState, setHudState] = useState(null);
   const [showMemoryLog, setShowMemoryLog] = useState(false);
+  const [brainConfig, setBrainConfig] = useState({ model: 'loading...', hasToken: false, localPipeline: false });
   const canvasRef = useRef(null);
 
+  useEffect(() => {
+    const fetchConfig = async () => {
+      try {
+        const response = await fetch('/api/config');
+        if (!response.ok) return;
+        const data = await response.json();
+        setBrainConfig({
+          model: data.model || 'unknown',
+          hasToken: Boolean(data.hasToken),
+          localPipeline: Boolean(data.localPipeline),
+        });
+      } catch (error) {
+        console.warn('Failed to load backend config:', error);
+      }
+    };
+    fetchConfig();
+  }, []);
+
   const addLog = (message, type = 'info') => {
     gameRef.current.logs = [{ id: Date.now(), message, type, time: Date.now() }];
-  };
+  }; 
 
   const startGame = (mode) => {
     WORLD_OBJECTS = generateWorld();
@@ -473,7 +490,7 @@ export default function App() {
     if (s.player.speedMult < 1) s.player.speedMult = Math.min(1, s.player.speedMult + 0.01);
   };
 
-  // ─── HUGGING FACE API CALL (replaces Anthropic callClaudeAPI) ────────────
+  // ─── BACKEND INFERENCE CALL (via /api/infer) ─────────────────────────────
   const callClaudeAPI = async (s) => {
     s.ai.llmThinking = true;
     const isNight = s.time < 6 || s.time > 18;
@@ -512,25 +529,13 @@ Valid Actions: FORAGE, HUNT, FISH, GET_WATER, SEEK_SHELTER, BUILD_CAMP, UPGRADE_
 Respond ONLY with a raw JSON object — no markdown, no extra text. Example: {"action":"FORAGE","thought":"Need wood and resources"} [/INST]`;
 
     try {
-      const response = await fetch(
-        `https://api-inference.huggingface.co/models/${HF_MODEL}`,
-        {
-          method: 'POST',
-          headers: {
-            'Authorization': `Bearer ${HF_TOKEN}`,
-            'Content-Type': 'application/json',
-          },
-          body: JSON.stringify({
-            inputs: prompt,
-            parameters: {
-              max_new_tokens: 80,
-              temperature: 0.7,
-              return_full_text: false,
-              stop: ['\n\n', '</s>', '[INST]'],
-            },
-          }),
-        }
-      );
+      const response = await fetch('/api/infer', {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+        },
+        body: JSON.stringify({ prompt }),
+      });
 
       if (!response.ok) {
         const err = await response.json().catch(() => ({}));
@@ -1173,12 +1178,7 @@ Respond ONLY with a raw JSON object — no markdown, no extra text. Example: {"a
         <div className="absolute inset-0 bg-[radial-gradient(ellipse_at_center,_var(--tw-gradient-stops))] from-blue-900/20 via-zinc-950 to-zinc-950"></div>
         <div className="max-w-3xl w-full space-y-6 z-10">
           <h1 className="text-5xl font-black tracking-tighter text-transparent bg-clip-text bg-gradient-to-r from-blue-400 to-emerald-400 border-b border-zinc-800/50 pb-6 flex items-center gap-4"><Brain className="text-blue-500" size={48}/> EVOLUTIONARY AI</h1>
-          <p className="leading-relaxed text-zinc-400 text-lg">Initialize Subject-01 into the high-fidelity 2.5D simulation. Powered by HuggingFace AI ({HF_MODEL || 'model not set'}) with persistent memory across 6 generations.</p>
-          {(!HF_TOKEN || !HF_MODEL) && (
-            <div className="bg-red-950/60 border border-red-700 rounded-xl px-5 py-3 text-red-300 text-sm font-mono">
-              ⚠ Missing env vars: {!HF_TOKEN ? 'VITE_HF_TOKEN ' : ''}{!HF_MODEL ? 'VITE_HF_MODEL' : ''}. AI will run in offline fallback mode.
-            </div>
-          )}
+          <p className="leading-relaxed text-zinc-400 text-lg">Initialize Subject-01 into the high-fidelity 2.5D simulation. Powered by HuggingFace AI with persistent memory across 6 generations.</p>
           <div className="grid grid-cols-2 gap-6 mt-8">
             <button onClick={() => startGame('hardcore')} className="bg-zinc-900/80 backdrop-blur border border-red-900/50 p-8 rounded-xl hover:bg-red-950/40 transition-all text-left flex flex-col gap-3 group shadow-2xl">
               <div className="flex items-center gap-3 text-red-500 font-bold text-2xl"><Skull size={28}/> HARDCORE MODE</div>