why transglot
Why translation belongs in your build, not in a spreadsheet.
Seven reasons, written to be checked rather than believed. Where a number does not exist, the row is left empty and labelled.
0707
reasons, counted as you scroll
the number fills as you read · scroll
what a language costs
The eleventh language costs what the first one did.
Per-word billing quietly deletes languages: every language you add is a line item somebody has to defend at the end of the month. An allowance does not work that way. A run only bills the rows it actually translated, so re-running a file nobody changed costs nothing.
cost per 1,000 words
$0.06
$0.80
$160
Team plan, inside the allowanceOverage, past the allowanceAll-human translation, industry anchor
log scale, because the bars span more than three decades and a linear axis would flatten two of them into the floor
Team is $49 a month for an 1,000,000 word allowance and overage meters at $0.0008 a word, both read from the published plan limits. The all-human bar is the $0.16 a word figure the ROI calculator ships as a reader-adjustable assumption, described there as a round industry rate for all-human translation rather than any vendor quote.
02
the gates
Nothing ships until four gates say yes.
Placeholders, plural forms, glossary terms and length, checked on every row of every run, on the draft pass and again on the review pass. A row that fails is repaired once and re-checked. If it fails again it is refused, not shipped.
placeholders
every token that went in comes back out
pluralsevery plural form the target language actually needs
glossarylocked terms held, never paraphrased
lengththe character ceiling set on the key, then repaired or refused
source · the QA pre-gate · retry bound · the quality gate settings
where the language lands
Translations land in the pull request, not in an email attachment.
The comparison here is with a workflow, not with a vendor. What changes is where the language ends up: in the same pull request as the code that needed it, or in a file travelling between two calendars.
| What changes | Transglotin the repository | The round tripexport, send, wait, import |
|---|---|---|
| Where the source strings live | The repository, read by the CLI or a connector | An export somebody made on a Tuesday |
| What starts a run | A push, on every commit | A person, when they remember |
| What checks the output | Four gates on every row, both passes | A reviewer, eventually |
| What a fix costs | A re-run of the changed rows only | Another round trip |
| Where a translation lands | The same pull request as the code | An attachment to merge by hand |
| What adding a language costs | One row on the project | A new quote |
Transglot
in the repository
- Where the source strings live
- The repository, read by the CLI or a connector
- What starts a run
- A push, on every commit
- What checks the output
- Four gates on every row, both passes
- What a fix costs
- A re-run of the changed rows only
- Where a translation lands
- The same pull request as the code
- What adding a language costs
- One row on the project
The round trip
export, send, wait, import
- Where the source strings live
- An export somebody made on a Tuesday
- What starts a run
- A person, when they remember
- What checks the output
- A reviewer, eventually
- What a fix costs
- Another round trip
- Where a translation lands
- An attachment to merge by hand
- What adding a language costs
- A new quote
No competitor is named, deliberately. A head-to-head figure would need a benchmark we have not run.
the tools you already use
It plugs into the 41 tools you already use.
Code hosts, content systems, storefronts, help desks, design files and file storage: 41 integrations, every one of them through the same four gates. Thirty-six are included on any paid plan and five start at Team.
SHShopify
WPWordPress
CFContentful
FGFigma
GHGitHub
GLGitLab
JRJira
NONotion
SASanity
SBStoryblok
WBWebflow
STStrapi
WCWooCommerce
MGMagento
BCBigCommerce
PSPrestaShop
ZDZendesk
ICIntercom
HSHelp Scout
DRDrupal
PRPrismic
DBDropbox
S3Amazon S3
GDGoogle Drive
+17→05
the meter
One price, one word allowance, and the overage rate on the page.
Per-word billing turns every new language into something somebody has to defend. An allowance does not, and the rate past it is printed here rather than discovered on an invoice.
illustrative · per-word billing
- 42,118 words at $0.16
- $6,738.88
- a second language, same month
- $6,738.88
- before anybody counts seats
- $13,477.76
Illustrative, priced at the $0.16 a word anchor the ROI calculator uses for all-human translation. It is a round industry rate, not a quote from any vendor.
transglot · Team
- words a month
- 1,000,000
- editor seats
- unlimited
- connectors
- all 41
- unchanged strings, re-run
- $0
$49/mo
Past the allowance, overage meters at $0.0008 a word, which is $80 per 100k words. Free is hard capped at 100,000 words to start and 10,000 a month after, and never meters at all.
06
Getting your translations out is one call.
There is no browser download of your translations. What exists is the read path your pipeline already uses every day: the API, the CLI, the delivery bundle and the storage sync jobs. All four need a project token, so they start at Starter.
documented exit
$ curl -H "Authorization: Bearer tgl_..." \
"https://api.transglot.ai/v1/pull?locale=de-DE&format=json"
200 · de-DE · json · etag set
# the same endpoint the CLI and the GitHub Action call
Deletion happens on request. The paperwork behind that sentence lives on the trust page rather than in this caption.
every claim is checkable
Check us before you trust us.
Every claim on this page is meant to be checkable in one click. Where a figure is not checkable, the row says so and stays empty.
what is public
MIT, on npm and on GitHub.
Our CLI is MIT on npm and our GitHub Action is MIT on GitHub, both public today. The hosted platform is a private application.
MIT
npm · transglot
transglot/transglot-action
what we will not print
→The figures you will not find here.
No uptime percentage, because we run no public probe: the only availability numbers we state are the contractual SLA targets, 99.9% on Business and 99.95% on Enterprise. No certification we do not hold. No customer we cannot name. No fixed count of supported languages, because the validator accepts any BCP-47 tag ICU recognises rather than a list.
- 01SLA targets are commitments, never measurements
- 02compliance status is read from config, never hardcoded
- 03708 ICU locales and 224 CLDR plural rule sets, both derived
the trust page has the receipts
FAQs
What people ask when they are comparing us with what they do today.
What does Transglot actually do?
It translates the strings in your product into every language you have added, automatically, and checks each row before it ships. Source files come in from your repository, a connector or an upload; translated files go back out through the API, the CLI or a delivery bundle. People review what they want to review, and nothing waits on them to start.
How is this different from using a translation agency?
An agency run starts when somebody exports a file and ends when somebody imports one back. Here the run starts on your commit and ends in the same pull request as the code. Adding a language is a row on the project rather than a new quote, and a fix is a re-run of the rows that changed rather than another round trip.
Why not just use a free machine translation API?
Raw machine translation has no idea what a placeholder is, which plural forms the target language needs, that your product name must not be translated, or that the string has to fit a button. Those four things are exactly what our gates check on every row, and a row that fails is repaired once and then refused rather than shipped.
Can I keep my own translators?
Yes. They work in the same grid with their own accounts, and a grant can be for one project and one language, so a freelancer sees French and nothing else. There are four roles, owner, admin, developer and translator, and all four exist on every plan including Free.
Do I still need a human to review the translations?
For the strings that carry your brand, yes, and the product is built to make that the small pile rather than the whole pile. Quality Mode is a switch on the project, available from the Team plan: turn it on where that wording lives, and let everything else ship on the fast pass. A row that fails a check arrives for review with the rendered screen attached where a screenshot exists.
What do I have to do to set it up?
Create a project, point it at your source file or connect a repository, and add the languages you want. The first run drafts, checks and repairs itself without further input. Connecting from CI needs a project token, which is available from the Starter plan.
Will it work with the tools we already have?
There are 41 integrations covering code hosts, content systems, storefronts, help desks, design files and file storage, and 20 file formats that round-trip natively. Thirty-six are included on any paid plan and five start at Team.
Can I bring over what we have in another tool?
Yes, from Lokalise, Phrase, Crowdin, Transifex, Weblate and POEditor. Keys, locales and translations come across in one guided import, and they land as human-authored rows, so the strings you have already paid for are reused rather than re-translated.
What happens to our translations if we leave?
They stay yours, and they come out the same way they come out every day: the API pull endpoint, the CLI, the delivery bundle or a storage sync. There is no download in the browser. All four need a project token, so they start at Starter.
Do you train AI models on our content?
No. We do not train any models on your content, and the inference providers we call are engaged for inference only, under commercial terms, each one named on the trust page. The exact-match memory pool is shared and anonymous, and any project can opt out of it on any plan.
Are you SOC 2 certified?
Not yet, and we will not put a badge on the site until we are. The SOC 2 Type II audit is in progress: Type I readiness first, then the observation window. The trust page states the status of every framework we are working through, read from config rather than typed into a design.
Is there a free plan, and what is the catch?
Free is $0 forever: 100,000 words to start and 10,000 a month after, one project, 15 language rows including your source, three seats, no card. The catch is that it is a hard cap rather than a meter, so translation stops until the month resets, and it cannot mint a project token, so the API, CLI, Action, webhooks and MCP server start at Starter.
seven reasons, one afternoon
Reason 08 is trying it.
A first language that lands in the same pull request as the code that needed it, checked by the same four gates as every row after it.
no card · 100,000 words to start on Free, then 10,000 a month · the four gates run on every row