No per-minute fees. No API keys. No CUDA tax.

Pro ($9.99/month) and Studio ($14.99/month) are monthly subscriptions billed through Razorpay. In India, plans are billed in INR (Pro ₹899/month, Studio ₹1,399/month) - Indian cards and UPI can only be charged in rupees, so pick INR at checkout if your card was issued in India. No per-minute metering, no API keys. Subscriptions are cancellation-only and non-refundable - cancel anytime and you keep access through the end of the cycle you already paid for, then revert to the free tier.

No - all sales are final and purchases are non-refundable. The 15-day free Studio trial (no card required) lets you fully evaluate Voxmelt on your own hardware before you pay, so once a billing cycle is charged it is not refunded, whether it is your first payment or a renewal. Only cancellation is supported: it stops future renewals, you keep full access through the cycle you have already paid for, then your account reverts to the free tier with no further charges. Full details are in the Refund & Cancellation Policy at voxmelt.com/privacy#refund.

15 days of everything - the full Studio pipeline. After the trial, the free tier keeps 30 minutes/day of dictation, 10 AI cleanup runs/day on the starter tone of every template pack, 10 minutes/day of Read Aloud with 2 voices (Clara and Marcus in 3 free tones), the 6-language translator, models up to 4B, and 1 custom template slot. Compose and Voiceover work within the same daily limits. You are never locked out of your own transcripts.

No account is required to record. An account only matters for managing your Pro/Studio license across machines.

Pro is the everyday plan: unlimited dictation, unlimited AI runs on the everyday template set (cleanup vibes, chat replies, meeting recaps, note distilling, signal extraction, translation), all 4 Voiceover personas and 8 delivery tones with speed control and WAV export, the Compose workspace, models up to 14B, 5 custom templates, and priority email support. Studio is the power-user plan: everything in Pro, plus the AI Prompt Architect and Code Prompt packs built for Cursor and Copilot workflows, every model in the catalog up to the 70B flagships, 30 custom template slots, and a direct line to the maintainer.

Never. Recording, transcription, and AI cleanup all run on your GPU - no cloud calls, no telemetry, no API keys. The only network traffic is license checks, billing, and update checks, and none of it carries audio or text. Turn off WiFi and dictation keeps working. That is the whole point.

Yes. Cancel from Account → Billing in one click. Cancellation stops future renewals only - you keep full Pro/Studio access through the end of the billing cycle you already paid for, then your account reverts to the free tier.

Wispr Flow sends your voice to the cloud and charges $15/month. Voxmelt does the same job on your own GPU for $9.99/month, and your audio never leaves your machine. You pick the AI model that cleans your text, it keeps working with WiFi off, and you can check the privacy claim yourself: open any network monitor, dictate, and watch zero audio leave your PC.

Superwhisper offers cloud-tier processing that uploads your audio. Voxmelt is 100% local across all tiers - even the AI post-processing runs on your GPU via Ollama. You also get smart GPU orchestration (auto VRAM management between Whisper and the LLM), which Superwhisper lacks. Plus full Windows support from day one.

Yes. Otter AI is cloud-only and bills per minute. Dragon is expensive and legacy. Voxmelt gives you the same Whisper large-v3 model that powers most modern cloud transcription, plus a local LLM that rewrites your output - all on your own hardware, for a flat monthly fee with no per-minute meter.

Yes. Since audio and text never leave your machine, there is no data processor to sign a BAA with - the data simply never transmits. Lawyers, doctors, and researchers use Voxmelt specifically because the privacy guarantee is architectural (local processing), not contractual (a ToS promise that could change).

Measured, not claimed: on our published benchmark, Voxmelt's local Whisper large-v3 averaged 7.8 percent word error rate across five scripted real-world clips on an RTX 3080 Ti, including 3.1 percent on fast speech with filler words and on background noise, while transcribing 10 to 15 times faster than real time. The per-clip tables, the method, and the scoring script are all public on the benchmark page, and the 15-day trial lets you test it on your own voice, which is the only benchmark that really matters.

Any NVIDIA GPU with 6 GB+ VRAM (RTX 3060 and up). An 8 GB card runs the small Whisper model comfortably. A 12 GB card (RTX 3080 Ti, 4070) handles large-v3 plus a 4B LLM. A 24 GB card (RTX 4090) keeps both models warm simultaneously for instant responses. Workstation cards (A6000, L40S, A100) run the 70B models.