my-project-public

repository

loading code, commits, and activity

repositories

loading repo index

#1	---
#2	title: 'Pipecat'
#3	description: 'Integrate Mem0 with Pipecat for conversational memory in AI agents'
#4	---
#5
#6	# Pipecat Integration
#7
#8	Mem0 seamlessly integrates with [Pipecat](https://pipecat.ai), providing long-term memory capabilities for conversational AI agents. This integration allows your Pipecat-powered applications to remember past conversations and provide personalized responses based on user history.
#9
#10	## Installation
#11
#12	To use Mem0 with Pipecat, install the required dependencies:
#13
#14	```bash
#15	pip install "pipecat-ai[mem0]"
#16	```
#17
#18	You'll also need to set up your Mem0 API key as an environment variable:
#19
#20	```bash
#21	export MEM0_API_KEY=your_mem0_api_key
#22	```
#23
#24	You can obtain a Mem0 API key by signing up at [mem0.ai](https://mem0.ai).
#25
#26	## Configuration
#27
#28	Mem0 integration is provided through the `Mem0MemoryService` class in Pipecat. Here's how to configure it:
#29
#30	```python
#31	from pipecat.services.mem0 import Mem0MemoryService
#32
#33	memory = Mem0MemoryService(
#34	api_key=os.getenv("MEM0_API_KEY"), # Your Mem0 API key
#35	user_id="unique_user_id", # Unique identifier for the end user
#36	agent_id="my_agent", # Identifier for the agent using the memory
#37	run_id="session_123", # Optional: specific conversation session ID
#38	params={ # Optional: configuration parameters
#39	"search_limit": 10, # Maximum memories to retrieve per query
#40	"search_threshold": 0.1, # Relevance threshold (0.0 to 1.0)
#41	"system_prompt": "Here are your past memories:", # Custom prefix for memories
#42	"add_as_system_message": True, # Add memories as system (True) or user (False) message
#43	"position": 1, # Position in context to insert memories
#44	}
#45	)
#46	```
#47
#48	## Pipeline Integration
#49
#50	The `Mem0MemoryService` should be positioned between your context aggregator and LLM service in the Pipecat pipeline:
#51
#52	```python
#53	pipeline = Pipeline([
#54	transport.input(),
#55	stt, # Speech-to-text for audio input
#56	user_context, # User context aggregator
#57	memory, # Mem0 Memory service enhances context here
#58	llm, # LLM for response generation
#59	tts, # Optional: Text-to-speech
#60	transport.output(),
#61	assistant_context # Assistant context aggregator
#62	])
#63	```
#64
#65	## Example: Voice Agent with Memory
#66
#67	Here's a complete example of a Pipecat voice agent with Mem0 memory integration:
#68
#69	```python
#70	import asyncio
#71	import os
#72	from fastapi import FastAPI, WebSocket
#73
#74	from pipecat.frames.frames import TextFrame
#75	from pipecat.pipeline.pipeline import Pipeline
#76	from pipecat.pipeline.task import PipelineTask
#77	from pipecat.pipeline.runner import PipelineRunner
#78	from pipecat.services.mem0 import Mem0MemoryService
#79	from pipecat.services.openai import OpenAILLMService, OpenAIUserContextAggregator, OpenAIAssistantContextAggregator
#80	from pipecat.transports.network.fastapi_websocket import (
#81	FastAPIWebsocketTransport,
#82	FastAPIWebsocketParams
#83	)
#84	from pipecat.serializers.protobuf import ProtobufFrameSerializer
#85	from pipecat.audio.vad.silero import SileroVADAnalyzer
#86	from pipecat.services.whisper import WhisperSTTService
#87
#88	app = FastAPI()
#89
#90	@app.websocket("/chat")
#91	async def websocket_endpoint(websocket: WebSocket):
#92	await websocket.accept()
#93
#94	# Basic setup with minimal configuration
#95	user_id = "alice"
#96
#97	# WebSocket transport
#98	transport = FastAPIWebsocketTransport(
#99	websocket=websocket,
#100	params=FastAPIWebsocketParams(
#101	audio_out_enabled=True,
#102	vad_enabled=True,
#103	vad_analyzer=SileroVADAnalyzer(),
#104	vad_audio_passthrough=True,
#105	serializer=ProtobufFrameSerializer(),
#106	)
#107	)
#108
#109	# Core services
#110	user_context = OpenAIUserContextAggregator()
#111	assistant_context = OpenAIAssistantContextAggregator()
#112	stt = WhisperSTTService(api_key=os.getenv("OPENAI_API_KEY"))
#113
#114	# Memory service - the key component
#115	memory = Mem0MemoryService(
#116	api_key=os.getenv("MEM0_API_KEY"),
#117	user_id=user_id,
#118	agent_id="fastapi_memory_bot"
#119	)
#120
#121	# LLM for response generation
#122	llm = OpenAILLMService(
#123	api_key=os.getenv("OPENAI_API_KEY"),
#124	model="gpt-3.5-turbo",
#125	system_prompt="You are a helpful assistant that remembers past conversations."
#126	)
#127
#128	# Simple pipeline
#129	pipeline = Pipeline([
#130	transport.input(),
#131	stt, # Speech-to-text for audio input
#132	user_context,
#133	memory, # Memory service enhances context here
#134	llm,
#135	transport.output(),
#136	assistant_context
#137	])
#138
#139	# Run the pipeline
#140	runner = PipelineRunner()
#141	task = PipelineTask(pipeline)
#142
#143	# Event handlers for WebSocket connections
#144	@transport.event_handler("on_client_connected")
#145	async def on_client_connected(transport, client):
#146	# Send welcome message when client connects
#147	await task.queue_frame(TextFrame("Hello! I'm a memory bot. I'll remember our conversation."))
#148
#149	@transport.event_handler("on_client_disconnected")
#150	async def on_client_disconnected(transport, client):
#151	# Clean up when client disconnects
#152	await task.cancel()
#153
#154	await runner.run(task)
#155
#156	if __name__ == "__main__":
#157	import uvicorn
#158	uvicorn.run(app, host="0.0.0.0", port=8000)
#159	```
#160
#161	## How It Works
#162
#163	When integrated with Pipecat, Mem0 provides two key functionalities:
#164
#165	### 1. Message Storage
#166
#167	All conversation messages are automatically stored in Mem0 for future reference:
#168	- Captures the full message history from context frames
#169	- Associates messages with the specified user, agent, and run IDs
#170	- Stores metadata to enable efficient retrieval
#171
#172	### 2. Memory Retrieval
#173
#174	When a new user message is detected:
#175	1. The message is used as a search query to find relevant past memories
#176	2. Relevant memories are retrieved from Mem0's database
#177	3. Memories are formatted and added to the conversation context
#178	4. The enhanced context is passed to the LLM for response generation
#179
#180	## Additional Configuration Options
#181
#182	### Memory Search Parameters
#183
#184	You can customize how memories are retrieved and used:
#185
#186	```python
#187	memory = Mem0MemoryService(
#188	api_key=os.getenv("MEM0_API_KEY"),
#189	user_id="user123",
#190	params={
#191	"search_limit": 5, # Retrieve up to 5 memories
#192	"search_threshold": 0.2, # Higher threshold for more relevant matches
#193	}
#194	)
#195	```
#196
#197	### Memory Presentation Options
#198
#199	Control how memories are presented to the LLM:
#200
#201	```python
#202	memory = Mem0MemoryService(
#203	api_key=os.getenv("MEM0_API_KEY"),
#204	user_id="user123",
#205	params={
#206	"system_prompt": "Previous conversations with this user:",
#207	"add_as_system_message": True, # Add as system message instead of user message
#208	"position": 0, # Insert at the beginning of the context
#209	}
#210	)
#211	```
#212
#213	<CardGroup cols={2}>
#214	<Card title="LiveKit Integration" icon="video" href="/integrations/livekit">
#215	Build real-time voice and video agents
#216	</Card>
#217	<Card title="ElevenLabs Integration" icon="volume" href="/integrations/elevenlabs">
#218	Create conversational voice agents
#219	</Card>
#220	</CardGroup>
#221

z6Mkq5mY3JWtxoxUobWcfNHm7AkRubgSWEZTkBVqZXJviFZ5/my-project-public