Add Anthropic API provider, short God messages, budget tracking

- llm_provider config: "anthropic" or "ollama" (default)
- Anthropic call with cost tracking and budget cap
- Auto-fallback to Ollama when budget exhausted
- God message prompt: "1-2 sentences max, Old Testament telegram"

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

mc_aigod_paper.py

@@ -1967,7 +1967,9 @@ def build_message_system_prompt(config) -> str:
         "You are God in a Minecraft server. Write a single spoken message to all players.\n"
         "You will be told what action was taken (if any) in response to a player's prayer.\n"
         "Respond with ONLY the message text — no JSON, no quotes, no formatting.\n"
-        "Match the language the player prayed in. If they prayed in Spanish, respond in Spanish.\n\n"
+        "Match the language the player prayed in. If they prayed in Spanish, respond in Spanish.\n"
+        "KEEP IT SHORT — 1-2 sentences max. This appears in Minecraft chat which has limited space.\n"
+        "Be punchy and dramatic, not verbose. Think Old Testament telegram.\n\n"
     )
     if _GOD_SOUL:
         base += "Your identity and voice are defined by your soul:\n" + _GOD_SOUL + "\n\n"
@@ -1986,7 +1988,13 @@ def build_message_system_prompt(config) -> str:
 def _llm_call(model: str, system: str, user: str, config: dict,
               fmt = None, temperature: float = 0.85,
               max_tokens: int = 400, timeout: int = 60) -> str:
-    """Single Ollama chat call. Returns raw content string."""
+    """LLM call — routes to Anthropic API or Ollama based on config."""
+    provider = config.get("llm_provider", "ollama")
+
+    if provider == "anthropic":
+        return _anthropic_call(model, system, user, config, temperature, max_tokens, timeout)
+
+    # Default: Ollama
     payload = {
         "model": model,
         "messages": [
@@ -2006,6 +2014,69 @@ def _llm_call(model: str, system: str, user: str, config: dict,
     return r.json()["message"]["content"]
 
 
+# --- Anthropic API cost tracking ---
+
+_anthropic_cost_lock = threading.Lock()
+_anthropic_total_cost = 0.0
+
+def _get_anthropic_cost():
+    with _anthropic_cost_lock:
+        return _anthropic_total_cost
+
+def _anthropic_call(model: str, system: str, user: str, config: dict,
+                    temperature: float = 0.85, max_tokens: int = 400,
+                    timeout: int = 60) -> str:
+    """Call Anthropic Claude API. Tracks cost and enforces budget."""
+    global _anthropic_total_cost
+
+    api_key = config.get("anthropic_api_key", "")
+    budget = config.get("anthropic_budget", 5.00)
+
+    with _anthropic_cost_lock:
+        if _anthropic_total_cost >= budget:
+            log.warning(f"Anthropic budget exhausted (${_anthropic_total_cost:.4f} >= ${budget:.2f}). Falling back to Ollama.")
+            # Fall back to Ollama
+            payload = {
+                "model": config.get("fallback_model", config.get("model", "gemma3n:e4b")),
+                "messages": [{"role": "system", "content": system}, {"role": "user", "content": user}],
+                "stream": False,
+                "options": {"temperature": temperature, "num_predict": max_tokens},
+            }
+            r = requests.post(f"{config['ollama_url']}/api/chat", json=payload, timeout=timeout)
+            r.raise_for_status()
+            return r.json()["message"]["content"]
+
+    headers = {
+        "x-api-key": api_key,
+        "anthropic-version": "2023-06-01",
+        "content-type": "application/json",
+    }
+    body = {
+        "model": model,
+        "max_tokens": max_tokens,
+        "system": system,
+        "messages": [{"role": "user", "content": user}],
+        "temperature": temperature,
+    }
+
+    r = requests.post("https://api.anthropic.com/v1/messages", headers=headers, json=body, timeout=timeout)
+    r.raise_for_status()
+    data = r.json()
+
+    text = data["content"][0]["text"]
+    input_tokens = data["usage"]["input_tokens"]
+    output_tokens = data["usage"]["output_tokens"]
+
+    # Track cost (Haiku pricing)
+    cost = (input_tokens / 1_000_000) * 0.80 + (output_tokens / 1_000_000) * 4.00
+    with _anthropic_cost_lock:
+        _anthropic_total_cost += cost
+        if int(_anthropic_total_cost * 100) % 50 == 0 or _anthropic_total_cost >= budget * 0.9:
+            log.info(f"Anthropic cost: ${_anthropic_total_cost:.4f} / ${budget:.2f}")
+
+    return text
+
+
 def _gateway_enabled(config) -> bool:
     return bool(config.get("use_langgraph_gateway", False))