Skip to main content

​
Overview

The synap-pipecat integration connects Synap, Maximem AI’s long-term memory layer, to Pipecat voice pipelines. It ships two FrameProcessor classes: SynapMemoryProcessor injects relevant user memory into the shared LLMContext before the LLM replies, and SynapRecorder captures each user and assistant turn back to Synap so memory persists across sessions. A SynapShortTermProcessor is also provided for injecting short-term conversation context.

Source Repository

Source code, examples, and issues for the Synap SDK and its Pipecat integration (under packages/integrations/synap-pipecat)

Documentation

Synap documentation from Maximem AI

Website

Learn more about Synap and request an API key

Dashboard

Manage your Synap account and API keys

​
Installation

This is a community-maintained package distributed separately from pipecat-ai: 
pip install maximem-synap-pipecat
Requires pipecat-ai>=0.0.80 and maximem-synap>=0.2.0.

​
Prerequisites

​
Synap Account Setup

Before using the Synap integration, you need:
  1. Synap Account / API Key: Get one at maximem.ai/synap
  2. A configured MaximemSynapSDK instance, created with your API key: MaximemSynapSDK(api_key="sk-...")

​
Configuration

​
SynapMemoryProcessor

Inserted before the user context aggregator. On each TranscriptionFrame, fetches user-scoped context from Synap and appends it as a system message to the shared LLMContext.
​
sdk
MaximemSynapSDK
required
A configured MaximemSynapSDK instance.
​
user_id
str
required
Required — Synap memory is user-scoped.
​
customer_id
str
default:"\"\""
Optional customer/org scope. An empty string means customer-less.
​
context
LLMContext
default:"None"
Shared LLMContext used by the user/assistant aggregators. When None, the processor is inert (no injection) but still passes frames through.
​
mode
str
default:"\"accurate\""
Synap fetch mode ("accurate" or "fast").
​
max_results
int
default:"10"
Cap on sdk.fetch results per turn.
​
include_conversation_context
bool
default:"False"
Whether to request Synap’s recent conversation block alongside the semantic memory.

​
SynapRecorder

Inserted after the assistant context aggregator. Buffers the latest user transcription and the streamed assistant response, then records both turns via sdk.conversation.record_message on LLMFullResponseEndFrame.
​
sdk
MaximemSynapSDK
required
A configured MaximemSynapSDK instance.
​
user_id
str
required
Required — Synap conversations are user-scoped.
​
customer_id
str
default:"\"\""
Optional customer/org scope. An empty string means customer-less.
​
conversation_id
str
default:"None"
Stable id for this call. Auto-generated per processor lifetime when absent.
A SynapShortTermProcessor is also exported for injecting short-term conversation context. See the source repository for its full configuration.

​
Usage

from pipecat.pipeline.pipeline import Pipeline
from pipecat.processors.aggregators.llm_context import LLMContext
from pipecat.processors.aggregators.llm_response_universal import (
    LLMUserAggregator,
    LLMAssistantAggregator,
)
from maximem_synap import MaximemSynapSDK
from synap_pipecat import SynapMemoryProcessor, SynapRecorder

sdk = MaximemSynapSDK(api_key="sk-...")
context = LLMContext(
    messages=[{"role": "system", "content": "You are a helpful assistant."}]
)

user_agg = LLMUserAggregator(context=context)
asst_agg = LLMAssistantAggregator(context=context)

pipeline = Pipeline([
    transport.input(),
    stt,
    SynapMemoryProcessor(sdk, user_id="alice", customer_id="acme", context=context),
    user_agg,
    llm,
    tts,
    transport.output(),
    asst_agg,
    SynapRecorder(sdk, user_id="alice", customer_id="acme"),
])
Reads degrade gracefully — if a Synap fetch fails, the frame continues downstream with no context injected so a Synap blip never breaks a live call. Write failures surface as a SynapIntegrationError pushed upstream as an ErrorFrame.

​
Compatibility

The integration requires pipecat-ai>=0.0.80. Check the source repository for the latest tested version and changelog.