待翻译：Show HN: LymeScribe – one computer on your network transcribes for the rest

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Mac & Windows · the server is in the room Transcription that stays on your hardware. For you, or your whole office. LymeScribe runs Whisper-class speech models on a machine you own — your own computer for a handful of people, or a dedicated box for the office. Not a cloud you rent. The audio never leaves the building. Download free Host or Server? Free as a client · paid hosting on your own computer or a dedicated office server · nothing uploaded, ever Recording… ↓ Any app · your cursor Follow up with Dana about the Q3 numbers before Friday's stand-up. on screen while you dictate · your words land in the app you're already in your mic → your machine → your transcript one person → a few people → the whole office audio in → text out → 0 bytes uploaded no account → no telemetry → no subscription your mic → your machine → your transcript one person → a few people → the whole office audio in → text out → 0 bytes uploaded no account → no telemetry → no subscription One promise, two doors Both run on hardware you own. Pick the one that fits the room. The free app connects to any LymeScribe server — that's the on-ramp. When you're ready to host, there are exactly two ways, and the only question is how many people you're serving. Host it yourself · for you & a few Your own computer doubles as the server. You and a handful of people on your network — off the computer you already own. A paid unlock in the same app — no extra hardware Embedded Whisper — runs on your CPU, GPU-accelerated when you have one — serves ~3 peers on your LAN One-time purchase, runs entirely on-device See pricing Server · for the office · Technology Preview A dedicated machine for the whole office. An always-on box that serves your whole team — built to scale past a few peers. GPU-accelerated — real-time transcription at office speed Built for many concurrent users, no peer cap Windows/NVIDIA or Apple Silicon, on your LAN Talk to us Why own the machine When the server is yours, the rules are yours. The popular dictation tools are genuinely good — and genuinely cloud-hosted. Every word crosses the internet to a server you don't control, governed by terms you didn't write and they can change. The price is monthly rent for as long as you keep talking. Owning the hardware flips every line of that deal. The audio never leaves the building, so privacy stops being a policy you have to trust and becomes a fact of where the machine sits. No account. No per-seat bill. Nothing to keep paying for. Modern hardware made this practical. A recent Mac or Windows PC runs Whisper-class models right on the CPU — a GPU (Apple Metal or NVIDIA CUDA) just makes the bigger models real-time-fast. So a computer you already own can be the server, and a dedicated box can serve the whole office. Either way, it's your hardware. Cloud dictation LymeScribe Where audio goes Their servers Never leaves your computer Pricing $8–15/mo, forever $39, once Account required Yes No Works offline No Completely Who owns the transcripts Read the ToS You. They're files on your disk. If the company disappears App stops working Nothing changes What it does A serious dictation tool that happens to respect you. 01 Push-to-talk, anywhere Tucks into the menu bar on Mac, the system tray on Windows. Hit a hotkey, speak, and the text lands in whatever app you're in — keystrokes or clipboard paste, configurable per app. 02 Whisper, on your hardware Runs OpenAI's Whisper models natively on the computer you own — the same model family the big cloud services build on, so the accuracy comes with it. Metal on Mac and CUDA on Windows make the larger models real-time-fast. No API, no per-minute fees. 03 Your history, your disk Every transcription saved locally, searchable, exportable. Not synced to anyone's cloud unless you put it there yourself. 04 Corrections dictionary Teach it your jargon, names, and acronyms once. Corrections apply automatically — and stay on your machine like everything else. 05 Host it yourself: your computer is the server A hosting unlock turns the Mac or PC you already own into a private transcription server for a few trusted devices on your network. One good machine serves them all. 06 Server: scale to the office LymeScribe Server runs on your own Windows/NVIDIA or Apple Silicon box — GPU-accelerated for real-time speed, serving many people at once on your LAN with no peer cap. Not just dictation Drop a recording in. Get back who said what. Dictation is only half the product. Drag any audio file — a meeting, an interview, a voicemail — onto LymeScribe and it transcribes the whole recording as a batch job, on your hardware, while you keep working. Speaker labeling tells the voices apart. Diarization runs alongside transcription, so a one-hour meeting comes back as a readable script — each line attributed to the person who said it — instead of a wall of undifferentiated text. Play a snippet of each voice back and type the real name once: "Speaker 1" becomes Maya everywhere in the transcript. Searchable, exportable, saved to your disk. Batch transcription and speaker labeling are part of hosting, and like everything else they run entirely on your machine — the recording never leaves the building. Speaker labeling wants a capable machine; see system requirements. What to expect: on Mac, Apple Silicon M2 or later with 16 GB+ memory is recommended — earlier or 8 GB machines work, but expect much longer processing times. On Windows, speaker labels require an NVIDIA GPU; transcript-only works on any 64-bit PC, though CPU-only processing is slow (a one-hour meeting can take several hours). team-standup.m4a24 min drop audio here done · 2 speakers · saved to your disk Speaker 1 Where did we land on the launch date — are we still good for the 14th? Speaker 2 Engineering's ready. I want QA to sign off before we announce anything. Speaker 1 Fair. QA signs off Friday, then it's official. Speaker 1 → Pricing Start free. Host when you're ready. Free as a client, forever. The two paid tiers are both about hosting — the only difference is scale. Client The on-ramp. Connect to any LymeScribe server. $0 free forever Connect to any LymeScribe server on your network Full dictation experience — hotkeys, history, corrections No account, no trial clock, no catch Download Early-adopter price Host Your own computer doubles as the server. You and a few people. $59$39 one-time · no subscription Host transcription on your own computer — fully on-device Serve yourself + a few devices on your network Pays for itself in about four months vs. $10/mo cloud dictation 7-day free trial, full features — no credit card 30-day money-back guarantee Free updates; discounted major upgrades later Start free trial Technology Preview Server A dedicated machine for your whole practice. Let's talk for offices & teams Always-on server for your whole practice — on your hardware GPU-accelerated for real-time speed at scale, no peer cap Batch jobs and speaker labeling for the whole team Windows/NVIDIA or Apple Silicon Contact us Early-adopter price Server A dedicated machine for your whole practice. $499$399 one-time · for offices & teams Always-on server for your whole practice — on your hardware GPU-accelerated for real-time speed at scale, no peer cap Batch jobs and speaker labeling for the whole team Windows/NVIDIA or Apple Silicon · 14-day free trial 30-day money-back guarantee Free updates; discounted major upgrades later Buy Server The Host unlock is purchased in-app after its 7-day trial; the Server license is purchased on the web after a 14-day trial. Payments handled by Paddle.com as merchant of record · 30-day refund policy Download Try it in the next two minutes. LymeScribe for Mac & Windows Free to use as a client. Hosting on your own computer starts a 7-day free trial — no account, no credit card. Download for Mac — .dmg Download for Windows — .msi MacmacOS 14+Apple SiliconNotarized by AppleAuto-updates via Sparkle WindowsWindows 10/1164-bitUnsigned installer · how to open past the security warning · auto-updates in-app System requirements Honest floors, stated up front. Local transcription is real compute. Rather than ship a degraded mode for aging machines, we publish the floor and stand behind everything above it. Below minimum, the app may install and even run — but performance there is outside what we support. Client free · connects to a server, which does the heavy lifting Minimum Recommended Mac macOS 14 · Apple Silicon or Intel Apple Silicon Windows Windows 10, 64-bit Windows 11 Host it yourself your own computer doubles as the server Minimum Recommended Mac Apple Silicon M2 · 8 GB unified memory M3 or newer · 16 GB or more Windows Windows 10, 64-bit · 4-core CPU · 8 GB RAM NVIDIA RTX GPU (CUDA) · 16 GB RAM Disk 5 GB free for models 10 GB free Below the Mac floor (Intel, M1), hosting runs in a CPU-only beta with the smaller models — it works, but the full large-model, real-time experience we stand behind starts at M2. On Windows, hosting runs on any modern CPU; an NVIDIA GPU is what makes the larger, more accurate models real-time. Speaker-labeled transcripts diarization · part of hosting, runs as a batch job Minimum Recommended Mac Apple Silicon M2 · 16 GB unified memory M-series Pro / Max / Ultra · 32 GB Windows NVIDIA GPU with 8 GB VRAM RTX 4070-class or newer, 12 GB+ VRAM Disk 10 GB free · one-time ~2 GB model download 10 GB free On Windows, speaker labeling requires the NVIDIA GPU — we don't offer it on CPU, where it's too slow to stand behind. (The office Server product can queue batch jobs on CPU; see its note below.) Server a dedicated machine for the whole office · Technology Preview Minimum Recommended Windows Windows 11 with WSL2 · NVIDIA GPU, 8 GB VRAM RTX 4070-class or better, 12 GB+ VRAM Mac Apple Silicon M2 · 16 GB unified memory M-series Pro / Max / Ultra · 32 GB No NVIDIA GPU? The server falls back to CPU for batch jobs — it works, but expect several times real-time. We support real-time performance only on the GPU configurations above. Questions Fair questions, straight answers. Host it yourself or Server — which do I need? Host it yourself if you want your own Mac or PC to do the hosting for yourself and a few people on your network — no extra hardware, a one-time unlock in the app. Server (Technology Preview) if you're equipping a whole office and want an always-on, GPU-accelerated machine built to serve many people at once. Both run on hardware you own; the only question is how many people you're serving. Is it really 100% offline? Transcription is — audio, models, and transcripts never leave your machine, and dictation works with Wi-Fi off entirely. The only network calls the app ever makes are checking for updates, and (for paid hosting) a license activation with our licensing server. Neither involves your audio or transcripts. How accurate is it? LymeScribe runs OpenAI's Whisper family of models — the same open models many cloud dictation services build on. The accuracy is in the model, not the subscription: on clear audio the larger models are competitive with the best cloud tools. When you host, you pick the model size your hardware can carry — bigger model, better accuracy — and the corrections dictionary teaches it your names, jargon, and acronyms so repeat mistakes get fixed once, permanently. What data do you collect? We never see your audio or transcripts — that's the entire point of the product. There are no analytics and no telemetry. When you buy a license, we store what's needed to issue and support it: your name and email, your license key, and basic activation records (device name, app version, and a non-reversible device identifier). Payment details go to Paddle, our merchant of record, not to us. Details in t [truncated for AI cost control]