Getting started

Quickstart

This starts where the installer leaves you: the app, open for the first time. By the end one node and one document are in, and you can find both again.

For the desktop app, start with installation. To use the account graph in a terminal, follow CLI key setup (the local graph needs no account key); signing in to the desktop app does not authorise terminal commands.

1. Sign in

The first launch fills the window with Sign in to Vinc. Choose Sign in with email or Continue with Google, finish signing in in the browser that opens, and you land back in the app. There is no close button, because nothing works before you sign in.

Once you are signed in you see your account graph. It is the same graph as the web app, so web assistants such as claude.ai and your other devices use it too. To keep it on this computer instead, turn on local mode.

If this computer already holds a graph, a screen tells you so and that graph stays in use.

Signed in in the browser, but the app is still waiting? In 0.9.6, the first sign-in screen has no route to Settings or manual key entry. When the app reports a failed or expired attempt, use the same sign-in button to try again. If it still cannot connect, contact support.

The desktop app does not read VINC_API_KEY. That environment variable is for the CLI; setting it does not sign you in to the app.

2. Write one node

An empty graph shows an invitation. Press New node.

The New node sheet opens. Fill in Title and Kind, and add One line and Topic if you like. The kind is one of four: Concept if you are writing what something means, Record for what you observed, Decision for what was chosen, Episode for something that happened on a day.

A test before you choose. If you can finish “X is …” in two sentences, it is a concept. If you cannot, there is more than one thing in there and it is worth splitting. What each kind holds is on the What Vinc is page.

3. Bring in one document

Press Choose a file on the empty-graph invitation, or drop a file onto it. You can also open the Tools icon at the right of the top bar and take the Import tab. Documents can be .md, .txt, .rst or .pdf.

A document keeps its full text and is split into chunks for search, which is why search lands on a passage rather than on a whole file.

Bringing a document in does not create concepts by itself. When the import finishes, a N concept candidates found · Apply button appears. Review the candidates and only the ones you pick become nodes. You can also open the document later and pull them out there.

4. Find it again

Type a word into the search field in the top bar. Search uses meaning and keywords together, so a near miss on the word still lands if the meaning is close. The list shows nodes, and Enter moves the View to the first result.

To find passages inside documents, open the Search screen (shortcut 2). Pressing a result raises the document column on the right with that passage in it. The column does not take width beside the View. It floats over it, pinned to the right edge.

When nothing matches, the app offers Open View: spread it out and look rather than ask.

5. Meet the default agents

A new graph is not empty. When you create an account or a team, the Default Agents pack comes in by itself: four roles chosen to fit a graph about anything.

RoleWhat it doesTier
agent-steward
Graph steward
Watches conventions, duplicates and topics, and proposes merges and retopics with evidence. Never applies them.Knowledge
agent-researcher
Research analyst
Turns research into a document, its concepts and an episode. Never takes the decision.Knowledge
agent-curator
Document curator
Takes documents in: stores the source, previews extraction, links mentions and proposes one topic per document.Execute
agent-reviewer
Claim reviewer
Checks every claim against the graph and labels it cited, assumed or unsourced. Never edits.Verify

The roles are concept nodes under the agent topic, listed under Agent roles on the left of the app. There is deliberately no orchestrator role handing out work.

In a host connected over MCP, each role shows up as a prompt (for example agent-reviewer). Using roles inside the app is covered in Using AI.

A role is a node in your graph, so change it to suit you. The documents it reads first, its checks and its paths start empty; fill them in for your own work. What a role is lives on the Agent page, and how to write a new one is in Writing a role.

Next

Attach an AI assistant. In the desktop app, Settings > AI connections registers Claude Desktop, Claude Code, Cursor and Codex CLI one at a time with Connect. MCP has the detail.
See every surface. Using the app covers each part of the window and everything inside it.
Why it was built this way. The Book answers in thirty-five chapters.