vinc-langgraph

Vinc for LangGraph and LangChain agents. Vinc is a knowledge graph you share with AI: the decisions, records and documents your team wrote, with the reasons attached. This package brings the relevant part of it into your graph nodes and model calls.

Install

pip install vinc-langgraph            # node helpers and tools
pip install "vinc-langgraph[agents]"  # plus the middleware for create_agent

Set VINC_API_KEY to a member key. A vinc_ro_ key is enough for everything except record_episode.

In a graph node

from vinc_client import VincClient
from vinc_langgraph import build_system_message, get_vinc_context

vinc = VincClient()

def review(state):
    block = get_vinc_context(vinc, state["messages"])   # newest human message
    system = build_system_message(block, "You review design changes.")
    return {"messages": [llm.invoke([system, *state["messages"]])]}

get_vinc_context makes one call, two when nothing matched or several nodes tied (it then widens with a search). It returns None when nothing matched, and also when Vinc is unreachable or the daily limit is spent, so the node goes on without it. A wrong key or space raises.

In a create_agent agent

from langchain.agents import create_agent
from vinc_langgraph import VincContextMiddleware

agent = create_agent("openai:gpt-5", middleware=[VincContextMiddleware()])

The middleware adds the block to the system message of each model call. An agent that uses tools calls the model several times per question, so the answer is kept for 60 seconds (cache_seconds) instead of spending a Vinc call each time.

Tools the model can call

from vinc_langgraph import create_vinc_tools

tools = create_vinc_tools(VincClient())   # vinc_brief and vinc_search, read only

Recording, after a person approved

Nothing in this package writes on its own. Put record_episode in the node after an interrupt(), and give only that node a vinc_sk_ key. examples/approve_then_record.py is a complete graph:

context -> review -> approve (interrupt) -> record

License

MIT