Deploy divination-ai from b75ed1c

README.md

@@ -0,0 +1,35 @@
+---
+title: Divination AI
+emoji: 🔮
+colorFrom: purple
+colorTo: indigo
+sdk: gradio
+sdk_version: "6.17.3"
+app_file: app.py
+pinned: false
+---
+
+# Divination AI
+
+An AI oracle that holds a brief conversation about your life situation, grounds its insights in real-world trends, and reveals 3 differentiated paths forward — always empowering, never deterministic.
+
+## How it works
+
+1. The Oracle asks 5 questions about your situation, goals, and strengths
+2. It searches the web for real-world context relevant to your domain
+3. MiniCPM4-8B generates exactly 3 scenarios (Momentum / Pivot / Leap)
+4. Each scenario includes a concrete first step, signals to watch for, and tradeoffs
+
+## Models
+
+| Role | Model | Params |
+|------|-------|--------|
+| Oracle (conversation + generation) | openbmb/MiniCPM4-8B | 8B |
+
+## Architecture
+
+- **Frontend**: Gradio 6.x on HuggingFace Spaces
+- **Backend**: Modal.com GPU services (A10G)
+- **Grounding**: Tavily web search API
+- **Structured output**: vLLM + xgrammar constrained decoding
+# Dark-themed scroll UI

model_runtime.py

@@ -0,0 +1,76 @@
+"""Divination AI — runtime client for the Modal backend."""
+from __future__ import annotations
+
+import os
+
+import httpx
+
+API_URL = os.environ.get(
+    "DIVINATION_AI_API_URL",
+    "https://rafalbogusdxc--divination-ai-api.modal.run",
+).rstrip("/")
+
+TIMEOUT_S = 900
+
+
+class BackendError(RuntimeError):
+    pass
+
+
+def health() -> dict:
+    try:
+        resp = httpx.get(f"{API_URL}/health", timeout=10, follow_redirects=True)
+        resp.raise_for_status()
+        return resp.json()
+    except Exception:
+        return {"status": "unreachable"}
+
+
+def intake(messages: list[dict]) -> dict:
+    try:
+        resp = httpx.post(
+            f"{API_URL}/intake",
+            json={"messages": messages},
+            timeout=TIMEOUT_S,
+            follow_redirects=True,
+        )
+        resp.raise_for_status()
+        return resp.json()
+    except httpx.ConnectError as e:
+        raise BackendError("Cannot reach the Oracle backend — is the Modal app deployed?") from e
+    except httpx.ReadTimeout as e:
+        raise BackendError("The Oracle is taking too long to respond (cold start). Try again in ~1 minute.") from e
+    except httpx.HTTPStatusError as e:
+        raise BackendError(f"Backend error {e.response.status_code}: {e.response.text[:300]}") from e
+
+
+def generate(messages: list[dict]) -> dict:
+    try:
+        resp = httpx.post(
+            f"{API_URL}/generate",
+            json={"messages": messages},
+            timeout=TIMEOUT_S,
+            follow_redirects=True,
+        )
+        resp.raise_for_status()
+        return resp.json()
+    except httpx.ConnectError as e:
+        raise BackendError("Cannot reach the Oracle backend — is the Modal app deployed?") from e
+    except httpx.ReadTimeout as e:
+        raise BackendError("The Oracle is taking too long (generation in progress). Please wait...") from e
+    except httpx.HTTPStatusError as e:
+        raise BackendError(f"Backend error {e.response.status_code}: {e.response.text[:300]}") from e
+
+
+def share_trace(messages: list[dict]) -> dict:
+    try:
+        resp = httpx.post(
+            f"{API_URL}/share",
+            json={"messages": messages},
+            timeout=30,
+            follow_redirects=True,
+        )
+        resp.raise_for_status()
+        return resp.json()
+    except Exception:
+        return {"shared": False, "reason": "Network error"}

requirements.txt

@@ -0,0 +1 @@
+httpx>=0.27