Data onboarding platform — governed, self-hostable, agent-ready

A data onboarding platform takes the files your customers actually have — CSV, Excel, JSON, XML, PDFs, scans — and turns them into clean, validated records inside your product. The category was defined by cloud services (Flatfile, OneSchema); Mildport is the self-hostable one: the same embeddable experience, with the engine, the data and the audit trail inside your boundary.

What “onboarding” has to cover

1. Every format customers bring — spreadsheets in five dialects, exports with renamed headers, a PDF invoice, a phone photo of a delivery note. One wizard, one result shape.
2. Mapping that explains itself — visible match scores, deterministic re-runs, learned per-tenant aliases your team can inspect and reset.
3. Validation before delivery — typed columns, required fields, review with diffs.
4. Delivery you can trust — signed webhooks with retries, a decision log per import, and receipts an auditor can verify.

Governed by design

Mildport treats an import as a governed protocol between three actors — the deterministic engine, optional AI, and a human authority:

• Ambiguities become typed questions with evidence, answerable by a person or (with the right entitlement) an AI agent — nothing silently guessed.
• Decisions carry provenance; applied imports can produce signed receipts verifiable offline, without access to your live system.
• Agents get the same governed surface over MCP — running inside your deployment, not a vendor cloud.

The full account of that design — all seven artifacts, and what each one lets you check — is on the import protocol page.

Where it runs

Anywhere you do: self-hosted with Docker Compose or Helm, an air-gapped all-in-one image, or fully in-browser for the zero-backend evaluation path. Pricing is published — hosted tiers and flat-annual self-host — on the pricing page.

See it work: try the playground — messy files included, no signup. Then the Quickstart embeds it in ~10 minutes.