> ## Documentation Index
> Fetch the complete documentation index at: https://agents.nanonets.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Translate Text

> Translates a list of short strings (field values, addresses, notes) into a target language, in input order, each flagged as translated or unchanged.

Translates a list of short strings — field values, addresses, notes, line-item descriptions — into a target language with the deployment's translation engines. Display name **"Translate Text"**. Off by default.

It is the text half of document translation. The review viewer's **Show Translation** overlay translates a document *page* (OCR, geometry, cache); this tool translates the strings you hand it, with no document, no cache and no geometry. Use it to show extracted values in the reviewer's language, not to translate a file.

## Authentication and enablement

No integration. Two gates, both must pass:

* The tool is enabled on the agent (off by default).
* The agent's **document translation** setting (`document_translation_enabled`) is on. Off is reported as a tool error, never as a silent pass-through. The agent's `document_translation_model` override picks the engine, exactly as it does for the page overlay.

Available to an agent step, to the Platform MCP, and to an **App Tool** bundle through `/run` (declare `{"op": "…", "tool": "translate_text", "params": ["texts", "target_language"]}` in the binding's `tools`; no configured instance is needed because there is nothing to configure).

## Inputs

* `texts` — strings to translate, in the order you want them back. At most **200 strings** and **20,000 characters** per call (characters, not bytes).
* `target_language` — BCP-47 code (`en`, `fr`, `de`, …). Defaults to `en`.

## Output

```json theme={null}
{
  "target_language": "en",
  "translations": [
    { "text": "Invoice", "translated": true },
    { "text": "", "translated": false }
  ],
  "translated_count": 1
}
```

`translations` is index-aligned with `texts`. `translated: false` means the engine handed the source back (already in the target language, or empty), so render it as untranslated rather than as a translation.

## Billing

One unit per string **actually translated** (`translated_count`), at the `translate_text` rate. Strings returned unchanged cost nothing. From an agent step the units land on that step; from an App Tool bundle they land on the `app_tool_run` step the call records, priced from the same rate.

## Limits and side effects

* Read-only; no data is stored. Identical strings in one call are deduplicated by the engine.
* Bounded per call as above; split larger sets across calls. From a bundle, **batch**: one call carrying every field of a form is one run (and, with `"record": false` on the op, no feed card), where one call per field is one run each.
* Engine selection and failover (Cloud Translation ↔ LLM) follow the deployment's `DOCUMENT_TRANSLATION_BACKEND` configuration.

## Expected errors

* Empty `texts`, more than 200 strings, or more than 20,000 characters — tool error naming the bound.
* Document translation not enabled for the agent — tool error (`not enabled`).
* Engine refused (quota, transport) — tool error carrying the engine's reason.
* No translation engine configured on the deployment — the step fails with `document translation is not configured`; from a bundle, `/run` answers `501`.