examples: add Synap persistent memory integration

examples/other/synap_memory.py

@@ -0,0 +1,85 @@
+"""Synap memory integration for LiveKit Agents.
+
+Demonstrates three patterns for giving a LiveKit voice agent persistent,
+cross-session memory via Synap (https://maximem.ai):
+
+1. preload_synap_context    — inject long-term memory before the call starts
+2. attach_synap_recording   — record every turn automatically
+3. synap_search_tool /
+   synap_store_tool         — LLM-callable tools for explicit memory search and store
+
+Install: pip install maximem-synap-livekit-agents
+Get an API key at synap.maximem.ai
+Open source: https://github.com/maximem-ai/maximem_synap_sdk/tree/main/packages/integrations
+"""
+
+import os
+
+from dotenv import load_dotenv
+from maximem_synap import MaximemSynapSDK
+from synap_livekit_agents import (
+    attach_synap_recording,
+    preload_synap_context,
+    synap_search_tool,
+    synap_store_tool,
+)
+
+from livekit.agents import Agent, AgentSession, ChatContext, JobContext
+from livekit.plugins import deepgram, openai, silero
+
+load_dotenv()
+
+sdk = MaximemSynapSDK(api_key=os.environ["SYNAP_API_KEY"])
+
+# In production, derive USER_ID from the room participant identity or JWT claims
+# so each caller gets their own Synap memory namespace.
+USER_ID = "demo-user-001"
+CUSTOMER_ID = "demo-customer"
+
+
+async def entrypoint(ctx: JobContext):
+    await ctx.connect()
+
+    chat_ctx = ChatContext()
+
+    # 1. Inject long-term memory before the first LLM turn
+    await preload_synap_context(
+        chat_ctx,
+        sdk,
+        user_id=USER_ID,
+        customer_id=CUSTOMER_ID,
+    )
+
+    agent = Agent(
+        instructions=(
+            "You are a helpful voice assistant with long-term memory. "
+            "Use synap_search to recall user preferences when relevant. "
+            "Use synap_store to save important new facts the user mentions."
+        ),
+        chat_ctx=chat_ctx,
+        tools=[
+            synap_search_tool(sdk=sdk, user_id=USER_ID, customer_id=CUSTOMER_ID),
+            synap_store_tool(sdk=sdk, user_id=USER_ID, customer_id=CUSTOMER_ID),
+        ],
+    )
+
+    session = AgentSession(
+        llm=openai.LLM(model="gpt-4o"),
+        stt=deepgram.STT(),
+        tts=openai.TTS(),
+        vad=silero.VAD.load(),
+    )
+
+    # 2. Automatically record every user transcription and assistant response
+    attach_synap_recording(session, sdk, user_id=USER_ID, customer_id=CUSTOMER_ID)
+
+    await session.start(
+        agent=agent,
+        room=ctx.room,
+    )
+
+
+if __name__ == "__main__":
+    from livekit.agents import WorkerOptions, cli
+
+    cli.run_app(WorkerOptions(entrypoint_fnc=entrypoint))