Translation shouldn't depend on your signal.
Lose your signal and most translators hand you a spinner. This one keeps its dictionaries and its offline models on your own disk, so an answer comes back.
macOS and Windows · Offline dictionaries in the free tier plus 10 offline model(s)
- lookup directions
- 72
- interface languages
- 10
- translation layers
- 5
- bytes uploaded (offline layers)
- 0
Five layers, and the network is the last one
It starts at the local cache and works down. An early hit costs you nothing. Plenty of sentences never reach layer five.
- L0
Local cache
Anything you've translated before comes back instantly. No network, no charge.
- L1
Offline dictionary
Lookups in 72 directions. Cantonese and the regional variants are in there too
- L2
Translation memory
Wording you corrected once gets reused
- L3
Offline model
Whole sentences, translated on your own machine, with the network off
- L4
AI translation
Only when the four layers above come up short
Every sentence shows its source
Each result carries the layer that produced it. One glance and you know whether that sentence touched the network at all.
What “offline-first” means here
Most tools treat offline as a failure state. No connection, so here's an error dialog. We built it the other way up. The cache is tried before the dictionary, the dictionary before your memory, memory before the offline model, and the network only gets asked at the point where everything already sitting on your own disk has been tried and come up short.
Where it earns its keep
At 35,000 feet
Ten hours, and a document in a language you don't read. In-flight Wi-Fi costs money and keeps dropping out. The dictionaries and the model are already on the laptop.
Behind the firewall
An air-gapped network can't reach outside services. The offline layers never need to, so the work carries on: everything those layers use was copied onto the machine at install time, and none of it expects a route out to exist.
No signal
Underground car park. Subway tunnel. A valley with one bar. You still get an answer back.
Sensitive material
Contracts, medical records, board papers, internal drafts. Whatever the offline layers handle stays on the machine.
The parts you'll use every day
Lookups that read like a dictionary
Type a word and you get a proper entry: pronunciation, part of speech, several senses, example sentences. Laid out the way a dictionary lays them out.
Whole documents
txt, Markdown, subtitles, Word, PDF. Formatting survives.
Translate a screenshot
Text baked into an image. An interface that won't let you select anything. An error dialog you can't copy out of. Grab it and read it.
Terminology on your terms
Import a glossary and your product names follow it exactly, or stay untranslated. Model numbers too. No fresh invention on every run.
Regional Chinese, properly
The characters convert, and so does the vocabulary: Taiwan and Hong Kong say different things about the same object. Cantonese is treated as its own language, because it is one.
Your wording sticks
Fix a translation once and it goes into memory. Same sentence next week, same wording, without you doing anything. People tend to notice this one a month in rather than on day one, which is a shame, because it's the feature that saves the most time.
Guides
How the five-layer chain works, how to get more out of the offline dictionaries and glossaries, and which setting controls what.
Understanding Quotas in JoesanTrans: What Costs Credits and What Remains Forever Free
This guide explains exactly which translation actions consume your daily AI character quota and why the offline-first architecture keeps four processing layers completely free, helping you maximize value without unexpected charges.
Which Content Leaves Your Computer, and What Doesn’t
This guide maps out JoesanTrans’s data flow, clarifying what remains strictly local, what triggers cloud requests, and how tier boundaries shape your daily translation workflow.
Screenshot Translation: Applicable Scenarios and Limitations
This guide clarifies when and how to leverage JoesanTrans for screen-captured text, explaining its local-first translation pipeline, inherent constraints, and optimal usage strategies across subscription plans.
Install it before you need it.
The free tier needs no account. Dictionaries and translation memory work the minute the installer finishes, and you can download offline models in the directions you pick.