diff --git a/adolf/openmemory/server.py b/adolf/openmemory/server.py
new file mode 100644
index 0000000..fd85b37
--- /dev/null
+++ b/adolf/openmemory/server.py
@@ -0,0 +1,62 @@
+import os
+from mcp.server.fastmcp import FastMCP
+from mem0 import Memory
+
+OLLAMA_CPU_URL = os.getenv("OLLAMA_CPU_URL", "http://host.docker.internal:11435")
+QDRANT_HOST = os.getenv("QDRANT_HOST", "host.docker.internal")
+QDRANT_PORT = int(os.getenv("QDRANT_PORT", "6333"))
+
+config = {
+    "llm": {
+        "provider": "ollama",
+        "config": {
+            "model": "qwen2.5:1.5b",
+            "ollama_base_url": OLLAMA_CPU_URL,
+        },
+    },
+    "embedder": {
+        "provider": "ollama",
+        "config": {
+            "model": "nomic-embed-text",
+            "ollama_base_url": OLLAMA_CPU_URL,
+        },
+    },
+    "vector_store": {
+        "provider": "qdrant",
+        "config": {
+            "collection_name": "adolf_memories",
+            "embedding_model_dims": 768,
+            "host": QDRANT_HOST,
+            "port": QDRANT_PORT,
+        },
+    },
+}
+
+memory = Memory.from_config(config)
+
+mcp = FastMCP("openmemory", host="0.0.0.0", port=8765)
+
+
+@mcp.tool()
+def add_memory(text: str, user_id: str = "default") -> str:
+    """Store a memory for a user."""
+    result = memory.add(text, user_id=user_id)
+    return str(result)
+
+
+@mcp.tool()
+def search_memory(query: str, user_id: str = "default") -> str:
+    """Search memories for a user using semantic similarity."""
+    results = memory.search(query, user_id=user_id)
+    return str(results)
+
+
+@mcp.tool()
+def get_all_memories(user_id: str = "default") -> str:
+    """Get all stored memories for a user."""
+    results = memory.get_all(user_id=user_id)
+    return str(results)
+
+
+if __name__ == "__main__":
+    mcp.run(transport="sse")
diff --git a/adolf/potential-directions.md b/adolf/potential-directions.md
new file mode 100644
index 0000000..b3ae04e
--- /dev/null
+++ b/adolf/potential-directions.md
@@ -0,0 +1,13 @@
+# Potential Directions
+
+## CPU Extraction Model Candidates (mem0 / openmemory)
+
+Replacing `gemma3:1b` — documented JSON/structured output failures make it unreliable for mem0's extraction pipeline.
+
+| Rank | Model | Size | CPU speed | JSON reliability | Notes |
+|------|-------|------|-----------|-----------------|-------|
+| 1 | `qwen2.5:1.5b` | ~934 MB | 25–40 tok/s | Excellent | Best fit: fast + structured output, 18T token training |
+| 2 | `qwen2.5:3b` | ~1.9 GB | 15–25 tok/s | Excellent | Quality upgrade, same family |
+| 3 | `llama3.2:3b` | ~2 GB | 15–25 tok/s | Good | Highest IFEval score (77.4) in class |
+| 4 | `smollm2:1.7b` | ~1.1 GB | 25–35 tok/s | Moderate | Use temp=0; NuExtract-1.5-smol is fine-tuned variant |
+| 5 | `phi4-mini` | ~2.5 GB | 10–17 tok/s | Good | Function calling support, borderline CPU speed |
diff --git a/adolf/test_pipeline.py b/adolf/test_pipeline.py
index b541c65..edacf6b 100644
--- a/adolf/test_pipeline.py
+++ b/adolf/test_pipeline.py
@@ -133,14 +133,14 @@ try:
     status, body = get(f"{OLLAMA_CPU}/api/tags")
     models = [m["name"] for m in json.loads(body).get("models", [])]
     has_embed = any("nomic-embed-text" in m for m in models)
-    has_gemma = any("gemma3:1b" in m for m in models)
+    has_qwen = any("qwen2.5:1.5b" in m for m in models)
     report("CPU Ollama reachable", True, f"models: {models}")
     report("nomic-embed-text present on CPU Ollama", has_embed)
-    report("gemma3:1b present on CPU Ollama", has_gemma)
+    report("qwen2.5:1.5b present on CPU Ollama", has_qwen)
 except Exception as e:
     report("CPU Ollama reachable", False, str(e))
     report("nomic-embed-text present on CPU Ollama", False, "skipped")
-    report("gemma3:1b present on CPU Ollama", False, "skipped")
+    report("qwen2.5:1.5b present on CPU Ollama", False, "skipped")
 
 
 # ── 4. Qdrant ─────────────────────────────────────────────────────────────────