feat: add StateCore as a memory method

.gitignore

@@ -16,4 +16,6 @@ cognee/.cognee_system/
 cognee/.data_storage/
 *.db
 croissant_files
-.env
+.env
+.venv/
+*.tgz

agent.py

@@ -80,6 +80,9 @@ def _initialize_agent_by_type(self, agent_config, dataset_config):
         elif self._is_agent_type("agentmemory"):
             from methods.agentmemory import initialize_agentmemory_agent
             initialize_agentmemory_agent(self, agent_config)
+        elif self._is_agent_type("statecore"):
+            from methods.statecore import initialize_statecore_agent
+            initialize_statecore_agent(self, agent_config)
         elif self._is_agent_type("rag"):
             self._initialize_rag_agent(agent_config, dataset_config)
         else:
@@ -277,7 +280,7 @@ def send_message(self, message, memorizing=False, query_id=None, context_id=None
         # Route to appropriate agent handler based on agent type
         if 'Long_context_agent' in self.agent_name:
             return self._handle_long_context_agent(message, memorizing)
-        elif any(self._is_agent_type(agent_type) for agent_type in ["letta", "cognee", "mem0", "zep", "knowl", "agentmemory"]):
+        elif any(self._is_agent_type(agent_type) for agent_type in ["letta", "cognee", "mem0", "zep", "knowl", "agentmemory", "statecore"]):
             return self._handle_memory_agent(message, memorizing, query_id, context_id)
         elif self._is_agent_type("rag"):
             return self._handle_rag_agent(message, memorizing, query_id, context_id)
@@ -418,6 +421,9 @@ def _handle_memory_agent(self, message, memorizing, query_id, context_id):
         elif self._is_agent_type("agentmemory"):
             from methods.agentmemory import handle_agentmemory_agent
             return handle_agentmemory_agent(self, message, memorizing, query_id, context_id)
+        elif self._is_agent_type("statecore"):
+            from methods.statecore import handle_statecore_agent
+            return handle_statecore_agent(self, message, memorizing, query_id, context_id)
         else:
             raise NotImplementedError(f"Memory agent type not supported: {self.agent_name}")
 

configs/agent_conf/RAG_Agents/gpt-4o-mini/StateCore_gpt-4o-mini-digest.yaml

@@ -0,0 +1,15 @@
+# The LLM-assisted arm of the StateCore pair: writes are stored as events and the engine's
+# own distillation runs (extraction into supersession-tracked facts, semantic conflict
+# resolution). The distillation model is gpt-5-mini -- the engine's recommended model; its
+# runtime is operated with gpt-5-class models -- while the reader stays gpt-4o-mini like every
+# other row. All other values are copied from Simple_rag_bm25, matching the deterministic arm.
+# The delta between the two StateCore rows is the measured value of the engine's distillation.
+agent_name: Agentic_memory_statecore_digest
+model: gpt-4o-mini
+temperature: 0.7
+input_length_limit: 10000000
+buffer_length: 200
+output_dir: ./outputs/statecore-digest-gpt-4o-mini
+
+retrieve_num: 10
+statecore_digest: true

configs/agent_conf/RAG_Agents/gpt-4o-mini/StateCore_gpt-4o-mini.yaml

@@ -0,0 +1,14 @@
+# StateCore (github.com/yul761/StateCore), driven through its published MCP front end --
+# the wrapper spawns `npx -y statecore-mcp@0.6.0` itself, so there is no service to start.
+# Every value except the output_dir is copied from Simple_rag_bm25 so the row sits beside
+# the published baselines on the same terms: retrieve_num 10 is what BM25, Zep, Cognee,
+# HippoRAG-v2 and the embedding baselines use. The memory side makes zero model calls
+# (deterministic supersession + lexical retrieval); see methods/statecore.py.
+agent_name: Agentic_memory_statecore
+model: gpt-4o-mini
+temperature: 0.7
+input_length_limit: 10000000
+buffer_length: 200
+output_dir: ./outputs/statecore-gpt-4o-mini
+
+retrieve_num: 10