Your coding agent moves hardcoded strings to keys and translates them. Deterministic code checks every LLM output (placeholders, CLDR plurals, glossary, pixel width) and refuses anything broken or too long, with the exact reason, until the retry is valid. Nothing invalid reaches your catalogs.
> Add a checkout page, in every locale
"checkout.items": "{count, plural,
one {# item} other {# items}}"Your agent adds a feature and new keys.
glot.translate({
keys: ["checkout.items"],
locales: ["fr", "ja", "ar"]
})glot gives it live context: glossary, refs, neighbours.
Every string is checked by code, not by vibes.
Only the delta, with an updated lockfile.
Each step is a CLI command your coding agent can run too, through the MCP server.
Finds hardcoded UI strings, proposes keys from context and rewrites the code for next-intl or react-i18next.
glot i18nize --yesOnly the delta, with your glossary, file:line context and UI role. Every string validated, retried, then escalated.
glot translateValidates every catalog in CI: syntax, placeholders, plural categories, glossary and pixel width. Exit 1 on any error, annotations on the PR.
glot check --format githubRenders the running app per locale, shortens what overflows its box, re-renders until it fits.
glot fit --locales deThe LLM proposes. Code decides what gets written.
ICU, i18next and MessageFormat 2 parsing. Placeholders, tags and plural categories checked for every locale. Failures are retried, then escalated to a stronger model.
glot.lock hashes every source and target. Only the delta gets translated, and hand edits are never overwritten.
glossary.toml lives next to your code. Required translations, forbidden terms and do-not-translate words are enforced, not suggested.
Claude Code, Cursor and friends query keys, glossary and issues directly, and translate with file:line references and UI role.
Playwright renders every locale and flags overflow, truncation and RTL layout breaks before your users see them.
Anthropic, OpenAI or any OpenAI-compatible endpoint, including local models. Costs are shown per job and capped by a budget.
MIT licensed. CLI, MCP server, API and this dashboard. Self-host it, fork it, audit it.
Pixel budget
German is often a third longer than English. glot fit captures your running app in every locale, measures each text against its container, asks the LLM for a shorter variant that keeps the meaning and the placeholders, then renders again.
$ glot fit --locales deReal captures from the demo app, German locale.
In-app editing
The dev overlay finds the key behind the text you clicked, shows it in every locale with live validator issues, and saves to your catalog files. Hot reload does the rest.
They work and are tested, but they are newer than the core loop and may change. Some need the experimental self-hosted server.
The LLM translates the target back into your language literally, explains its choices and gives a verdict. Cached until the source or the target changes.
“Relance” is the usual French term for a payment reminder; the tone is slightly firmer.
Every locale of a page side by side, with overflow, truncation and RTL issues from visual QA pinned to the screen they break.
An anonymous beacon counts visitors by preferred language and the locale you served them, then compares conversion. No cookie, no visitor id.
Example data
For languages you don't ship yet, @glot/runtime translates in the browser with the built-in Translator API, then runs glot's validator compiled to WebAssembly.
Where glot differs from hosted translation platforms and other developer-first tools.
| Feature | glot | Lokalise | Crowdin | Lingo.dev | Tolgee |
|---|---|---|---|---|---|
| Price | Free (MIT) + your LLM usage | From $149/mo, 14-day trial | Paid plans after a 14-day trial, free for open source | Free CLI with your own key, hosted engine $0 base + usage | Free tier and free self-host, cloud from €58/mo |
| Open source | |||||
| Bring your own LLM key | |||||
| Deterministic validation of every string | |||||
| Git as source of truth (lockfile) | |||||
| MCP server for coding agents | |||||
| In-context editing in your app | |||||
| Rendered pixel-budget fitting |
Checked against each vendor's public website and docs on October 18, 2026. Partial means limited, plan-dependent or through an integration. No means we found no such feature in public docs. Plans change: every cell is linked to its source.
Run it locally with your agent, or connect it to a glot server for your team.
cargo install glotnpx glot initglot i18nize --yes{
"mcpServers": {
"glot": {
"command": "glot",
"args": [
"mcp"
],
"env": {
"GLOT_SERVER": "http://localhost:8790",
"GLOT_TOKEN": "<your-token>"
}
}
}
}