llama-index-memory-unison

Long-term agent memory for LlamaIndex — powered by the Unison brain.

Drop in wherever you'd use ChatMemoryBuffer. Agents get persistent, knowledge-graph-powered recall across sessions with zero infrastructure changes.

Install • Quick start • Replacing ChatMemoryBuffer • How it works • For agents

Reading this as an AI agent? See AGENTS.md — install, auth, and UnisonMemory.from_client(session_id=...) usage in one place.

Install

pip install llama-index-memory-unison

Quick start

import os
from llama_index.memory.unison import UnisonMemory
from llama_index.core.agent import ReActAgent
from llama_index.llms.openai import OpenAI

# Build Unison memory for this user/session
memory = UnisonMemory.from_client(
    session_id="user-123",          # stable per-user or per-thread ID
    api_key="usk_live_...",         # or set UNISON_TOKEN env var
    # api_url="https://brain.unisonlabs.ai",  # default
    # search_k=5,                             # hits per recall
)

# Pass it to any LlamaIndex agent or chat engine
llm = OpenAI(model="gpt-4o", api_key=os.environ["OPENAI_API_KEY"])
agent = ReActAgent.from_tools([], llm=llm, memory=memory, verbose=True)

# Subsequent calls automatically recall relevant long-term context
response = agent.chat("What did we discuss last week about Alice?")
print(response)

Replacing ChatMemoryBuffer

# Before
from llama_index.core.memory import ChatMemoryBuffer
memory = ChatMemoryBuffer.from_defaults(token_limit=3000)

# After — drop-in replacement with long-term recall
from llama_index.memory.unison import UnisonMemory
memory = UnisonMemory.from_client(session_id="user-123", api_key="usk_live_...")

Using environment variables

export UNISON_TOKEN="usk_live_..."
export UNISON_API_URL="https://brain.unisonlabs.ai"  # optional, this is the default

from llama_index.memory.unison import UnisonMemory

# Reads UNISON_TOKEN and UNISON_API_URL automatically
memory = UnisonMemory.from_defaults(session_id="user-123")

Environment variables

How it works

Network errors are caught and logged; they never crash the agent.

Releasing

Set a PyPI token, then run one command:

export TWINE_USERNAME=__token__
export TWINE_PASSWORD=pypi-...
python3 scripts/release.py

Or configure ~/.pypirc instead of the env vars.

The script builds llama-index-memory-unison, publishes to PyPI (idempotent — skips if the version is already on PyPI), then tags and pushes v<version>.

Links

• Unison brain: https://unisonlabs.ai
• Docs: https://docs.unisonlabs.ai
• Unison brain repo: https://github.com/unison-labs-ai/unison-brain
• LlamaIndex memory docs: https://docs.llamaindex.ai/en/stable/module_guides/storing/chat_stores/

Contributing & security

Contributions welcome — see CONTRIBUTING.md. Found a vulnerability? See SECURITY.md — please report privately, not via a public issue.

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.

License

MIT — see LICENSE.