Auto-Translation in Postbox — Every Submission in Your Language

You run a SaaS with a global audience. Your contact form gets submissions in Spanish, Japanese, German, Portuguese. Every time one comes in, you copy the message, open Google Translate, paste it, read the result, go back to your inbox. Multiply that by twenty submissions a day.

We built auto-translation in Postbox so you never have to do that again.

How it works

When you enable auto-translation on a form, every incoming submission is automatically translated to your language before it hits your inbox. The original data is preserved — we store the detected source language alongside the translation, so you always have access to exactly what was submitted.

The pipeline order matters. Spam detection runs first. Translation runs second. We don’t waste credits translating junk. If a submission is flagged as spam by our spam filtering layer, it never reaches the translation step. Your credits are spent only on legitimate submissions that you’ll actually read.

After translation, smart replies can draft a response based on the translated content — in your language, with full context. The entire processing pipeline is sequential and intentional.

Field-aware translation

This is the part that matters most, and the part no schemaless backend can replicate.

Postbox knows your form’s schema. It knows that message is a string field containing human-written text. It knows that email is an email address. It knows that name is a person’s name. So when a submission comes in, we translate the message field and leave the email and name fields untouched.

{
  "name": "Yuki Tanaka",
  "email": "yuki@example.com",
  "message": "御社のAPIドキュメントについて質問があります。"
}

After auto-translation:

{
  "name": "Yuki Tanaka",
  "email": "yuki@example.com",
  "message": "I have a question about your API documentation."
}

The name stays as-is. The email stays as-is. Only the field containing translatable text gets translated.

A schemaless form backend receiving that same payload has no idea which fields contain natural language and which are identifiers. It would have to guess. Does “Yuki Tanaka” need translating? Is “yuki@example.com” a sentence in some language it doesn’t recognize? Without a schema, every field is an opaque string.

We wrote about why this structural advantage matters across the entire processing pipeline in why structured forms beat schemaless endpoints. Translation is one of the clearest examples.

LLM-powered, tuned for accuracy

We use LLM-powered translation with a low temperature setting (0.2) to prioritize accuracy over creativity. This isn’t a use case where you want the model to get inventive with word choices. You want the closest faithful rendering of what the person wrote.

The source language is detected automatically from the content. We store it as metadata on the submission — so you can filter submissions by language, see patterns in where your global audience is coming from, and always refer back to the original text.

This is a meaningful upgrade over rule-based translation services. LLMs handle colloquialisms, context-dependent phrasing, and mixed-language input (code-switching) far better than traditional machine translation systems. A support message that mixes casual Japanese with English technical terms gets translated accurately, not mangled.

On-demand translation via MCP

Auto-translation runs automatically when enabled on a form. But what about submissions that came in before you turned it on? Or forms where you want translation selectively?

If you’re on the Pro plan with MCP access, you can trigger translation on-demand using the translate_submission tool. Point it at any submission, specify the target language, and it translates in place — same field-aware logic, same accuracy, same credit cost.

This is useful for workflows where an AI agent is triaging submissions and decides that a particular one needs translation. The agent calls the tool, the submission gets translated, and the agent continues processing. No human in the loop.

The use cases

Global SaaS support. Your product serves users in thirty countries. Support messages arrive in a dozen languages. Without auto-translation, your support team either needs to be multilingual or needs to context-switch into a translation tool for every non-English submission. With Postbox, every message arrives in your language. Response times drop.

International lead generation. You’re running landing pages in multiple markets. Contact form submissions come in the local language. Your sales team speaks English. Auto-translation means leads don’t sit unread because nobody on the team reads Korean. Every lead is actionable the moment it arrives.

Multilingual feedback collection. You ship a product survey. Responses come back in whatever language your users think in. Instead of batching translations manually at the end, every response is immediately readable. You can spot trends in real-time instead of waiting for someone to translate a spreadsheet.

Agent-driven triage. An AI agent monitors your submissions via MCP, categorizes them, and routes them to the right team. Auto-translation means the agent works with normalized, single-language data. No language detection logic in your agent code. No translation API integration to maintain. Postbox handles it before the agent ever sees the submission.

What it costs

Translation costs 1 AI credit per submission. Only submissions that are actually in a different language consume a credit — if someone submits in your language, no translation runs, no credit is spent.

On the Free plan, you get 50 lifetime AI credits. These are shared across all AI features — spam filtering, auto-translation, and smart replies. Once they’re gone, AI processing stops.

On Pro ($19/month), you get 500 AI credits per month that replenish every billing cycle. If you exceed 500, translations continue at $0.005 per submission — metered, not blocked. For context, that’s half a cent per translated submission. A thousand extra translations in a month costs five dollars.

Why this matters

The thesis behind Postbox is that the post-submit layer should be intelligent. Not a dumb pipe that stores whatever arrives. An endpoint that understands what it received, validates it, cleans it, and presents it in a form you can act on immediately.

Auto-translation is one of the clearest expressions of that idea. The raw submission is valuable. The translated submission is actionable. Postbox turns one into the other automatically, using the schema to do it accurately, and charges less than a penny for the privilege.

Your data arrives. You read it. You act on it. The language it was written in is no longer your problem.