plugin-pipecat

Documentation Index

Fetch the complete documentation index at: https://docs.mem0.ai/llms.txt Use this file to discover all available pages before exploring further.

Pipecat Neocortex Integration

Integrate Neocortex with Pipecat to add long‑term conversational memory to your voice and chat agents.

This package provides a NeocortexMemoryService that plugs into the Pipecat pipeline in the same way as Mem0MemoryService, but stores and retrieves memories from Neocortex instead.

Installation

Install from your workspace (or virtualenv) that runs Pipecat:

pip install neocortex-pipecat

You will also need to set your Neocortex API key as an environment variable:

export TINYHUMANS_API_KEY=your_neocortex_api_key

Optionally, configure a custom Neocortex base URL:

export TINYHUMANS_BASE_URL=https://api.your-backend.com

Configuration

Neocortex integration is provided through the NeocortexMemoryService class.

from neocortex_pipecat import NeocortexMemoryService, NeocortexParams

memory = NeocortexMemoryService(
    api_key=os.getenv("TINYHUMANS_API_KEY"),  # Your Neocortex token/JWT
    user_id="unique_user_id",                 # Unique identifier for the end user
    agent_id="my_agent",                      # Identifier for the agent using the memory
    run_id="session_123",                     # Optional: specific conversation session ID
    params=NeocortexParams(
        search_limit=10,                      # Max Neocortex memory chunks per query
        system_prompt="Here are your past memories:",
        add_as_system_message=True,           # Add memories as system (True) or user (False) message
    ),
)

• At least one of user_id, agent_id, or run_id must be provided.
• NeocortexParams.search_limit is passed to Neocortex as maxChunks.

Endpoint Mapping (Mastra-aligned)

This plugin now includes the same endpoint families exposed by the Mastra plugin:

• POST /v1/memory/insert
• POST /v1/memory/query
• POST /v1/memory/admin/delete
• POST /v1/memory/sync
• POST /v1/memory/recall
• POST /v1/memory/memories/recall
• POST /v1/memory/chat
• POST /v1/memory/interact
• POST /v1/memory/documents
• POST /v1/memory/documents/batch
• GET /v1/memory/documents
• GET /v1/memory/documents/:documentId
• DELETE /v1/memory/documents/:documentId
• POST /v1/memory/queries
• POST /v1/memory/conversations
• POST /v1/memory/interactions
• POST /v1/memory/memories/thoughts
• GET /v1/memory/ingestion/jobs/:jobId

Pipeline Integration

Place NeocortexMemoryService between your context aggregator and LLM in the Pipecat pipeline:

from pipecat.pipeline.pipeline import Pipeline

pipeline = Pipeline([
    transport.input(),
    stt,                # Speech-to-text for audio input
    user_context,       # User context aggregator
    memory,             # Neocortex memory service enhances context here
    llm,                # LLM for response generation
    tts,                # Optional: Text-to-speech
    transport.output(),
    assistant_context   # Assistant context aggregator
])
``]

## Advanced API Surface

`NeocortexMemoryService` now also exposes Mastra-style wrappers you can call directly:

- `sync_memory`
- `recall_memory_master`
- `recall_memories`
- `chat_memory`
- `interact_memory`
- `insert_document`
- `insert_documents_batch`
- `list_documents`
- `get_document`
- `delete_document`
- `query_memory_context`
- `chat_memory_context`
- `record_interactions`
- `recall_thoughts`
- `get_ingestion_job`
 
Run it with:

```bash
export TINYHUMANS_API_KEY=your_neocortex_api_key
export OPENAI_API_KEY=your_openai_key
uvicorn examples.voice_demo:app --reload --host 0.0.0.0 --port 8000

Then connect to /chat from your Pipecat-compatible client UI and speak; the agent will remember and reuse past conversation details via Neocortex.