Agent node
Write the agent as a node.
A role is a node in your graph with its mission, what it must never do and who it hands work to. Hosts and frameworks read it fresh each run, so an edit applies from the next one.
How roles workFor every moment
you work with knowledge.
A knowledge graph you share with AI.
Connect meaning.
Compose Agents.
In Vinc, people and AI build knowledge together, then compose new agents on top of it.
Your graph lives in your vincs.io account, the same on the web and on the desktop. To keep it on one machine, turn on local mode in Settings. How the builds differ
Yours, your team's, everyone's.
One graph, filled in together.
Personal
Knowledge that is yours alone.
In your account, or with local mode in a folder on your machine.
Start freeTeam
Knowledge you write with colleagues.
One shared graph with viewer, editor and admin roles. Everyone keeps their own personal graph.
Create a teamCommunity
Knowledge everyone uses.
Pages anyone can read and improve, each with its sources and its history.
Find a communityYours to keep. It stays where you put it, in a form you can take anywhere.
Keep it on this machine
Local mode keeps the graph in a folder you pick, with no copy in your account.
Your data is one folder
Copy the folder. That is the backup.
Sync only the topics you pick
Opt-in per topic. The rest never leaves the device.
A key that only reads
Hand one out and it cannot have written anything.
Nothing from the app is ever sent to us.Every feature in the docs
One graph, many ways in. See it, search it, and hand it to your assistant.
Your whole graph, in one view.
Concepts and the connections between them join into one graph.
Find it by word, or by meaning.
From the exact words to what you meant, reach what you need fast.
AI reads your graph.
One line in your client and the assistant you already use works from your own knowledge.
$ claude mcp add --transport http vinc https://mcp.vincs.io/mcp
Added HTTP MCP server vinc with URL: https://mcp.vincs.io/mcp to local config
> why do we generate CSS from tokens?
● vinc - vinc_search (MCP)(query: "CSS generated from tokens")
⎿ Tokens are the single source, CSS is generated · Decision
● You decided this in vinc/design: the tokens are the single source, and the CSS is generated from them.
Put AI to work on your own graph. Your graph becomes the workspace the AI works in.
Every agent reads the same graph. An assistant over MCP, a framework agent and your own script all reach the same nodes. None of them keeps a copy.
Choose how your agent works. Write it as a node, run it in a framework, or call the graph from your own code.
from langchain.agents import create_agentfrom vinc.integrations.langgraph import \ VincContextMiddleware role = "concept:agent-reviewer-judgement"agent = create_agent("openai:gpt-5", middleware=[VincContextMiddleware( prefetch_role=role)])
from agent_framework import Agentfrom vinc.integrations.agent_framework \ import VincContextProvider role = "concept:agent-reviewer-judgement"agent = Agent(client=chat_client, # any model context_providers=[VincContextProvider( prefetch_role=role)])
$ pip install "vincs[langgraph]"pip install "vincs[agent-framework]"
Agent framework
Run it in your framework.
Packages for LangGraph, LangChain and Microsoft Agent Framework load the role and the matching nodes before each model call, with ids the model can cite. Nothing is written back unless your code asks.
Framework guideSDK
Call it from your own code.
The vincs package carries the Python SDK and the vinc command. Give the reading parts a read-only key and keep the write key for the step that records.
Python SDK guideCommunity
Knowledge everyone keeps improving.
Read shared pages, follow their sources and help improve them.vincs.io/community · Anyone can read
Pricing
Personal Free costs nothing, for one person, forever. Team is $12 per seat per month.
See pricing