Skip to content
Skip to content

Prepare data from an on-premises system

How to send a reviewed export from your own systems to Compute without giving Lithi any remote access to them.

Updated

Make a controlled local export first

Your application makes the export inside your environment. Your agent keeps the task goal, context and final decision. When the selected workflow provides one, a compatible client or Kit helper prepares the declared encrypted artifact handoff; Lithi accepts the artifact reference after transfer permission, availability, quote and approval, then returns the admitted result for you to inspect before local import.

Use your existing database, file-share or application tools to make a bounded export inside your environment. Start with the data owner and the permitted processing boundary; if policy forbids external processing, stop before upload. An on-premises source does not make a remote Compute job local or air-gapped.

This guide does not install a database connector, open a firewall or give Lithi an SSH session. Keep source credentials in your environment, and disclose only the fields your policy allows.

Freeze and describe the export

Finish the export and stop its writers before preparing work. Retain the exact files, byte counts, content digests, export time and selected columns. Map each record to a stable item ID. A mutable path or live table name isn't a fixed input identity.

This Python example fingerprints synthetic bytes locally. It makes no network request and isn't an encryption or submission example.

import hashlib
import json
payload = b'{"item_id":"sample-001","text":"Example record"}\n'
record = {"bytes": len(payload), "sha256": hashlib.sha256(payload).hexdigest()}
assert record["bytes"] > 0 and len(record["sha256"]) == 64
print(json.dumps(record, sort_keys=True))

Hand off only the approved content

Inspect representative rows for private fields, duplicate IDs, missing values and size limits. Export a format accepted by the selected workflow. Keep customer mappings and sensitive source paths in your protected preparation records. A digest identifies bytes; it doesn't make sensitive content safe to disclose.

Follow file preparation for the authorized encryption and artifact handoff. The native job uses the artifact reference, not a local filesystem path or database password. Review current access, workflow availability and the quote before approving paid work. Lithi handles the admitted job; your local export process remains yours.

Verify the result before local import

Retrieve the accepted output and inspect exceptions. Verify its declared integrity, then use your own application to import reviewed rows into a separate staging area. Join by item ID, not row order. Approve any subsequent overwrite or application update separately.

If the export changes, prepare and review a new input. If a local import fails, retain the accepted result and repair that import without rerunning Compute. If access expires, renew only the needed permission. Remove temporary plaintext and local copies under your own retention policy; an expired download link isn't proof of deletion.

Next, check data suitability and processing boundaries.

Was this page helpful?