langchain-unison

LangChain memory, history & retriever — backed by the Unison brain.

Install • UnisonMemory • UnisonChatMessageHistory • UnisonRetriever • Env vars • For agents

🤖 Reading this as an AI agent? Jump to AGENTS.md — it covers install, auth (UNISON_TOKEN), the three classes, the recall-before / persist-after pattern, and the contributor guide, all in one page.

LangChain memory, chat history, and retriever backed by the Unison brain — a personal knowledge graph that stores and recalls context across every session, tool, and agent.

Install

pip install langchain-unison

Quick start

UnisonMemory

Drop-in memory for any LangChain chain. On every call it recalls relevant context from your brain and ingests the new turn when done.

from langchain_unison import UnisonMemory
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser

memory = UnisonMemory()  # reads UNISON_TOKEN from env

template = PromptTemplate.from_template(
    "Context from your knowledge base:\n{history}\n\nQuestion: {input}\nAnswer:"
)

# Load memory before the chain call
inputs = {"input": "What pricing model did we decide on?"}
inputs.update(memory.load_memory_variables(inputs))

# ... run your LLM chain with `inputs` ...
outputs = {"output": "We went with a freemium tier at $49/mo."}

# Save the exchange back to the brain
memory.save_context(inputs, outputs)

UnisonChatMessageHistory

BaseChatMessageHistory that appends every message to the Unison brain in real time.

from langchain_unison import UnisonChatMessageHistory
from langchain_core.messages import HumanMessage, AIMessage

history = UnisonChatMessageHistory(session_id="my-session-001")

history.add_message(HumanMessage(content="What is the refund policy?"))
history.add_message(AIMessage(content="Refunds are available within 30 days."))

print(history.messages)
# [HumanMessage(content='What is the refund policy?'),
#  AIMessage(content='Refunds are available within 30 days.')]

UnisonRetriever

BaseRetriever that queries GET /v1/brain/search and returns LangChain Document objects.

from langchain_unison import UnisonRetriever

retriever = UnisonRetriever(k=6)

docs = retriever.invoke("What did we decide about the enterprise tier?")
for doc in docs:
    print(doc.metadata["title"], "—", doc.page_content[:120])

Environment variables

You can also pass token= and base_url= directly to any class constructor.

For agents

See AGENTS.md — install, auth, usage patterns, and contributor guide in one page.

Links

• Homepage: https://unisonlabs.ai
• Docs: https://unisonlabs.ai/docs
• Unison brain repo: https://github.com/unison-labs-ai/unison-brain
• Issues: https://github.com/unison-labs-ai/langchain-unison/issues

Releasing

Set a PyPI token — either via ~/.pypirc or environment variables:

export TWINE_USERNAME=__token__
export TWINE_PASSWORD=pypi-...

Then run:

python3 scripts/release.py

This builds langchain-unison, publishes it to PyPI (idempotent — skips if the version is already on PyPI), then tags and pushes the release tag.

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.