Data import

Turn rows, records, and documents from 21 sources into markdown pages with frontmatter.

Every imported record becomes one markdown page. Fields land in frontmatter, so you can search, filter, and query them with DQL right away. Imports go through the normal write pipeline, so each page is versioned and indexed like any other edit.

Terminal

kiwifs import --from <source> [flags] --root ./knowledge

Supported sources

Databases

postgres, mysql, sqlite, mongodb, firestore, dynamodb, redis, elasticsearch

Files

csv, json, jsonl, yaml, excel, bibtex, markdown

Apps and wikis

notion, airtable, gsheets, obsidian, confluence

Croissant

croissant dataset metadata from Kaggle, Hugging Face, and others

Airbyte

Airbyte connectors through Docker or Airbyte Cloud

Documents

PDF, DOCX, PPTX, and more through MarkItDown

Where each source is available

The CLI supports every source. The REST API, the web UI import wizard, and the kiwi_import MCP tool support a subset.

Source	CLI	REST POST /import	MCP kiwi_import
postgres, mysql, sqlite, mongodb, firestore	Yes	Yes	Yes
csv, json, jsonl	Yes	Yes	Yes
notion, airtable	Yes	Yes	Yes
markdown, obsidian, bibtex, croissant	Yes	Yes	No
yaml, excel, gsheets, confluence	Yes	No	No
dynamodb, redis, elasticsearch	Yes	No	No
Airbyte connectors (firebase-rtdb, …)	No	Yes	No

REST and MCP imports read file and directory paths on the machine that runs KiwiFS, not on the caller’s machine. To send a file from your machine, use POST /api/kiwi/import/upload. The kiwi_import MCP tool only works in local mode (kiwifs mcp --root).

Common flags

These flags work with every source.

Flag	Default	Description
--from		Source type (required)
--root, -r	./knowledge	Knowledge root to write into
--prefix	Source name	Folder for the imported pages
--id-column	Auto-detect	Field whose value becomes the file name
--columns	All	Comma-separated fields to keep
--limit	0 (no limit)	Maximum records to import
--dry-run	false	Count what would be imported without writing

The default --prefix is the source’s name: the table or collection, the file name without its extension, the vault folder name, notion, or redis.

How records become pages

KiwiFS writes each record to <prefix>/<id>.md. The <id> is the --id-column value, the table’s primary key, or an id or _id field. Without any of those it falls back to the row number.

users/42.md

---
_imported_at: "2026-10-19T18:46:15Z"
_source: users
_source_id: pg:users:42
email: [e-mail]
name: Jane Doe
role: admin
---

# Jane Doe

| Field | Value |
|-------|-------|
| email | [e-mail] |
| name | Jane Doe |
| role | admin |

The page title comes from a title or name field, falling back to the ID. Sources that already produce markdown (markdown, obsidian, confluence, bibtex, croissant) keep their own body and frontmatter. KiwiFS adds the _source, _source_id, and _imported_at fields to it.

Re-imports

Imports are idempotent. When a page with the same _source_id exists and its fields are unchanged, KiwiFS skips it. Changed records are rewritten, and new records are created.

Output

Imported 12 records, skipped 40

Generate a schema from a sample

For csv, json, and jsonl files, --infer-schema samples up to 100 rows and prints a JSON Schema instead of importing. Add --save-schema to also write it to .kiwi/schemas/<name>.json under the current directory.

Terminal

kiwifs import --from csv --file people.csv --infer-schema --save-schema

See Schemas for how KiwiFS validates frontmatter against these files.

Related

Browse, preview, and sync

Preview rows, save connections, and schedule auto-sync.

Import API

REST reference for every import endpoint.