pipecat-unison

pipecat-unison

Your voice agent forgets the caller the second they hang up. Give it a memory.

A Unison brain integration for Pipecat voice agents.

Install · Usage · API · License

---

With Unison vs. without

Without Unison	With Unison
Caller says their allergy on call #1. Agent asks again on call #2.	UnisonPipecatService retrieves the allergy from the brain before the LLM speaks.
Every session starts from zero — no preferences, no history.	Relevant memories are injected into the system prompt on every turn.
Session data lives only in the audio log — inaccessible to the agent.	A session note is written to the brain after each user turn; knowledge compounds.
Memory is per-session, per-agent, per-tool.	One brain. All your agents read the same source of truth.

---

Powered by the Unison brain — not a flat vector store. Temporal facts that know what changed when, entity resolution that knows who's who, and one source of truth shared across every agent and teammate — voice, Claude Code, Cursor, Codex, your backend.

Why Unison, not a vector store bolted onto your voice agent (or mem0)?

Other memory	Unison
Stores what the caller said as a flat log / vector dump	Resolves who and what they meant and when it changed — a temporal knowledge graph
A silo — scoped to this one agent	One brain every agent and teammate reads from and writes back to
Greets a returning caller with a fact that's no longer true	Bitemporal supersession stops surfacing the version that's no longer true
"Trust our benchmark"	An open, reproducible benchmark scoring every system — including ours

---

Install

Prerequisites: Bun and Python ≥ 3.10.

# Frontend deps
bun install

# Backend deps
cd backend && python -m venv venv && source venv/bin/activate && pip install -r requirements.txt && cd ..

Configure:

cp backend/.env.example backend/.env   # fill in UNISON_TOKEN + OPENAI_API_KEY
cp .env.example .env                   # fill in VITE_UNISON_TOKEN (optional)

backend/.env:

UNISON_TOKEN=usk_live_...
OPENAI_API_KEY=sk-...

.env (optional — enables the in-browser memory panel):

VITE_UNISON_TOKEN=usk_live_...
VITE_BACKEND_URL=http://localhost:8001

---

Usage

# Terminal 1 — voice backend
cd backend && source venv/bin/activate && python server.py

# Terminal 2 — React frontend
bun run dev

Open http://localhost:5173, click the circle to start a voice session.

---

API

Drop UnisonPipecatService between the user aggregator and the LLM:

import os
from unison_pipecat import UnisonPipecatService
from pipecat.pipeline.pipeline import Pipeline

memory = UnisonPipecatService(
    token=os.getenv("UNISON_TOKEN"),       # or omit — reads env automatically
    base_url=os.getenv("UNISON_API_URL"),  # optional; defaults to https://brain.unisonlabs.ai
    user_id="user_123",
    session_id="session_abc",
    params=UnisonPipecatService.InputParams(
        mode="full",        # "profile" | "query" | "full"
        search_limit=10,
    ),
)

pipeline = Pipeline([
    transport.input(),
    stt,
    user_aggregator,
    memory,   # <-- inject here, between aggregator and LLM
    llm,
    tts,
    transport.output(),
    assistant_aggregator,
])

Parameters

Parameter	Default	Description
token	$UNISON_TOKEN	Unison API token (usk_live_...)
user_id	"default"	User identifier — namespaces brain documents
session_id	"default"	Session identifier — per-session note paths
base_url	$UNISON_API_URL	Override the Unison API base URL
params.mode	"full"	profile / query / full
params.search_limit	10	Number of search hits to retrieve (1–50)
params.search_threshold	0.0	Minimum relevance score to include
params.memory_tag	"pipecat"	Tag applied to written brain documents

Modes

• query — semantic + keyword search for the current user message
• profile — reads a maintained profile document at /private/notes/user-{user_id}-profile.md
• full — both (default, recommended)

Verify connectivity

from unison_pipecat import UnisonClient
import os

client = UnisonClient(token=os.environ["UNISON_TOKEN"])
print(client.whoami())   # {"user": {...}, "workspace": {...}, "scopes": ["brain:read","brain:write"]}
print(client.status())   # {"docCount": N, ...}

Get a token

# Provision a machine key
curl -X POST https://brain.unisonlabs.ai/v1/auth/provision \
  -H "Content-Type: application/json" \
  -d '{"email": "agent@example.com"}'
# returns: {"apiKey":"usk_live_..."}

# Verify with OTP from email (makes the token permanent)
curl -X POST https://brain.unisonlabs.ai/v1/auth/verify \
  -H "Content-Type: application/json" \
  -d '{"email": "agent@example.com", "code": "123456"}'

Or with the Unison CLI:

npx @unisonlabs/cli login

Environment variables

Variable	Required	Description
UNISON_TOKEN	Yes	Unison API token (usk_live_...)
UNISON_API_URL	No	Override API base URL (default: https://brain.unisonlabs.ai)
OPENAI_API_KEY	Yes (backend)	OpenAI key for STT / LLM / TTS
VITE_UNISON_TOKEN	No	Frontend token for the in-browser memory panel
VITE_BACKEND_URL	No	Override backend URL (default: http://localhost:8001)

---

Star history

---

License

MIT — see LICENSE.

---

Part of the Unison Labs constellation

One brain, every agent. Every repo below reads from and writes to the same Unison brain — no per-tool memory silos.

Repo	What it does
unison-brain	CLI · SDK · MCP server — the core
claude-unison	Memory for Claude Code
cursor-unison	Memory for Cursor
codex-unison	Memory for OpenAI Codex CLI
opencode-unison	Memory for OpenCode
openclaw-unison	Memory for OpenClaw
pipecat-unison	Memory for Pipecat voice agents ← you are here
python-sdk	Python SDK for the brain
install-mcp	One-command MCP installer
code-chunk	AST-aware code chunking
unison-fs	Mount the brain as a filesystem
backchannel	Async messaging between agents
Unison-evals	Open memory benchmark suite