File-based datasource plugin for @inferagraph/core. Loads nodes from delimited files (CSV/TSV) or a folder of Markdown files with frontmatter.
pnpm remove @inferagraph/file-datasource
pnpm add @inferagraph/fileThe class was also renamed FileDatasource → FileDataSource. Most hosts can switch to the new factory:
import { fileDataSource } from '@inferagraph/file';pnpm add @inferagraph/file @inferagraph/coreimport { fileDataSource } from '@inferagraph/file';
const datasource = fileDataSource({
type: 'csv',
path: './data/nodes.csv',
hasHeader: true, // first row contains column names
contentFields: ['summary'], // optional: surface columns via getContent
edges: [
{
id: 'e1',
sourceId: 'n1',
targetId: 'n2',
attributes: { type: 'related_to' },
},
],
});
await datasource.connect();
const view = await datasource.getInitialView();
const node = await datasource.getNode('n1');
const neighbors = await datasource.getNeighbors('n1');
const results = await datasource.search('keyword');
await datasource.disconnect();import { fileDataSource } from '@inferagraph/file';
const datasource = fileDataSource({
type: 'markdown',
path: './data/nodes', // folder containing .md files
frontmatter: ['id', 'name', 'type'], // required keys per file (exact match)
edges: [],
});
await datasource.connect();
// Each .md file becomes a node with frontmatter as attributes;
// the body is exposed via getContent.
const content = await datasource.getContent('n1');
// { nodeId: 'n1', content: '<body markdown>', contentType: 'markdown' }The factory is the preferred entry point, but the class is exported for hosts that need to subclass:
import { FileDataSource } from '@inferagraph/file';
class MyFileDataSource extends FileDataSource {
// ...
}| Option | Type | Description |
|---|---|---|
type |
'csv' | 'tsv' | 'markdown' |
File format |
path |
string |
File path (csv/tsv) or folder path (markdown) |
edges |
EdgeData[] |
Host-supplied relationships |
idField |
string |
Column / frontmatter key for the node id (default: 'id') |
| Option | Type | Description |
|---|---|---|
hasHeader |
boolean |
First row is column names (default: true) |
columns |
string[] |
Required when hasHeader=false. Length must match the parsed first row. |
contentFields |
string[] |
Columns surfaced via getContent as metadata. When unset, getContent returns undefined. |
| Option | Type | Description |
|---|---|---|
frontmatter |
string[] |
Required keys. Each .md file must contain EXACTLY this set (extras and missing keys throw). |
-
connect()reads and parses every file up-front (deterministic, fail-fast). -
disconnect()clears the parsed state. - All seven
DataAdaptermethods (getInitialView,getNode,getNeighbors,findPath,search,filter,getContent) require an active connection. - Edges supplied in
config.edgesare stored as-is and used for graph traversal.
MIT
