Skip to main content

Import data from CSV

Prompt an AI coding assistant to translate a CSV or TSV export into Infrahub object YAML. Your assistant writes the file, creates a branch, validates and loads it, and hands the branch back for you to review.

Prerequisite

Requires an AI coding assistant with the Infrahub Skills installed. Without the MCP server or infrahubctl configured, your assistant works from local schema files instead, which is less accurate about your live schema.

What it does​

Working from the Data Importer skill, your assistant takes a CSV or TSV export — one file, a folder, or one large denormalized sheet — and checks its columns against your schema before writing anything. It tries these sources in order and uses the first one available:

SourceRequiresWhat it sees
MCP serverThe MCP server connected to your instanceLive, branch-specific state — including dropdown choices and HFIDs added on a feature branch
infrahubctl schema exportinfrahubctl configured against your instanceThe live schema, not branch-specific unless you pass --branch
/api/schema REST endpointNetwork access to the instance's APIThe same live schema as the CLI export
Local schemas/*.yml filesNothing — works with no connection to any instanceWhatever is on disk, which can lag what's actually deployed

The rows are translated into the same object YAML format documented on Load data using YAML file.

Your assistant then checks its own output against the object-file rules and confirms every relationship reference it can — checking against the batch's own files and any live objects it already looked up — before it contacts a server. When only local schema files are available, it can confirm that references within the batch line up, but it can't tell whether a reference to an object that only exists live on the server is valid; that still needs a server-side check. Only once the local check passes does it create a fresh branch, run infrahubctl object validate against it to catch schema-resolution errors — including any reference that still doesn't resolve — and then infrahubctl object load to put the objects on the branch. Your assistant hands that branch back to you and never merges it.

Where a column maps to nothing in your schema, your assistant stops and asks rather than guessing.

Worked example​

1. Provide the file​

Example prompt
Import this inventory.csv into Infrahub.

2. Review the generated object YAML​

Your assistant maps each column to a schema attribute or relationship and writes one numbered object file per kind, each carrying a provenance comment:

objects/001-devices.yml
# Generated from inventory.csv
# sha256: <hash of inventory.csv>  generated: <UTC timestamp>
---
apiVersion: infrahub.app/v1
kind: Object
spec:
  kind: NetworkDevice
  data:
    - name: sw01-svc01
      site: Stonehenge Visitor Centre

3. Your assistant creates the branch, validates, and loads​

infrahubctl branch create csv-import-20260813-1004
infrahubctl object validate objects/ --branch csv-import-20260813-1004
infrahubctl object load objects/ --branch csv-import-20260813-1004

You do not run these yourself. Your assistant runs them in this order, and only after its own local check passes. infrahubctl object validate catches schema-resolution errors — a kind that does not exist, a reference that does not resolve — before anything loads. The branch is named csv-import-YYYYMMDD-HHMM by default.

4. Review the diff and open a Proposed Change​

Review the loaded objects in the web interface, then open a Proposed Change to merge the branch into main. See Proposed Changes for the review workflow.

What to watch for​

An unmapped column stops the import instead of guessing at a match. Resolve the mismatch — rename the column, or add the attribute to your schema — and run it again, rather than working around it. Check the branch name before merging; csv-import-YYYYMMDD-HHMM is the default, not a required format.

Loading is not transactional across files. If one file in a large batch fails partway through, the branch already has the earlier files applied. Delete the branch and run it again with a fresh name rather than repairing the partial state by hand.