Build your graph in Savanna

Savanna nests your work in two objects. A workgroup is a container for related workspaces and the boundary for access control; name it after a team or project. A workspace is the compute that runs your graph, and it’s what you start, stop, and resize.

When you first register, Savanna creates a workgroup and a workspace for you. Wait until the workspace status is active, then continue below. You do not need to create either one to try the product.

To add another project later, Create Workgroup is a two-step wizard: the workgroup, then its first workspace. Choose Read/Write so the workspace can load data and install queries (a Read-Only workspace can query an existing graph but not load into it). You can resize later without rebuilding.

Full screens: create a workgroup and create a workspace.

Open Design Schema from the navigation menu, the workspace page, or the GSQL Editor. The schema is the structure of your graph: which entities exist, how they connect, and what each one stores.

Start by creating a graph from the graph selector dropdown and giving it a name. Your first graph on a workspace can take up to two minutes while the services warm up.

Then lay out the model:

Give each vertex type a primary key that uniquely identifies an entity, since that’s what the loader matches on when it creates and deduplicates vertices. Under Advanced Settings you can index non-primary attributes you expect to filter on, and add a reverse edge on a directed edge when you need to traverse it in both directions.

Modeling reference: Schema Designer.

Open Load Data and select an active workspace, then pick your connector. Savanna ships step-by-step flows for local files, Amazon S3, Google Cloud Storage, and Azure Blob Storage, with Snowflake and JDBC alongside them. Connectors without a guided flow yet hand you a GSQL template to finish in the editor instead.

The guided flow moves through three decisions:

Configure the file. Savanna detects delimiters and line breaks and shows you the parsed result. If the columns split wrong, change the delimiter, end-of-line, or quoting options. An enclosing character such as a double quote overrides the delimiter inside a token, which is what you want when string values contain commas. You can rename header columns here to something readable, since those names are what you’ll map against.

Map columns to the schema. For each vertex and edge type, choose which source column feeds which attribute. Quick Map does the obvious matches in one pass: map all to target aligns existing attributes with matching headers, while map all from source also creates new attributes for headers that don’t match anything yet. Where the source doesn’t quite fit the target, a token function transforms the value as it loads.

Confirm and run. Savanna shows you Schema to be changed and Data to be loaded before anything executes. Read the schema diff carefully; some schema changes drop existing data, and the warning on that screen is the last checkpoint. Confirm to start the loading jobs, then watch their Status as they run.

Load vertices before the edges that reference them, so each edge finds both endpoints already present.

Open the GSQL Editor. Queries are written in GSQL, TigerGraph’s query language, and the editor gives you a file list on the left, the editing panel in the middle, results at the bottom, and the Schema Designer on the right when you need to check a type.

Create a query with the + next to your graph in the Query List. A useful first query starts from one vertex and walks one hop out: select a vertex by its primary key, traverse an edge type, and return the neighbors. That single traversal is the thing a graph database does that a relational join makes painful, so it’s worth writing by hand once.

A custom query has to be installed before it can run. Select it in the Query List and click Install, then run it and read the result panel. Expect vertices, edges, paths, or computed values depending on what your query returns.

For a visual read on the same data, Explore Graph lets you walk vertices and their neighbors, and pattern search finds paths between them without writing GSQL. It’s the fastest way to confirm your edges connect what you think they connect.

Editing, installing, and sharing queries: GSQL Editor guide.

If a loading job fails, the mapping is the usual cause: check that the parsed columns line up with the attributes you targeted and that primary keys aren’t empty in the source. Fix the mapping and rerun the job; you don’t need to recreate the workspace or the schema.