AI Services

Private audio transcription and LLM automation running on our own GPU hardware in Romania. Pay-as-you-go in EUR. Your data never leaves our infrastructure.

Privacy is the strongpoint

Most AI services route your data through OpenAI, Google, Anthropic, or Microsoft. Convenient, but the trade-off is real: your call recordings, your documents, your customer conversations all leave your premises and become part of someone else’s pipeline.

Custom Logisoft Labs runs its own AI infrastructure on local GPU hardware in Romania.

When you transcribe an audio file with us, the file is processed on our box, the result is sent back to you, and the audio is deleted. When you call our LLM, the prompt and the response never see a third-party API. We don’t train models on your data — we couldn’t, even if we wanted to.

That’s the foundation. Everything else is service.

What we offer

Whisper transcription

Production-grade speech-to-text built on OpenAI’s open-source Whisper models — running on our own GPU, exposed over a single REST endpoint:

  • Languages: 99 languages supported by Whisper large-v3 / large-v3-turbo
  • Formats: any audio file ffmpeg can read (MP3, WAV, M4A, OGG, …)
  • Telephony-aware: if you record caller and callee on separate channels, we stitch the two transcripts into one timestamped A/B conversation
  • Optional SRT output for direct subtitling workflows
  • Translate-to-English mode in addition to native-language transcription

Typical use cases: call-centre QA, meeting and interview transcription, podcast captioning, compliance archiving.

Qwen LLM automation

A private LLM API powered by Qwen3-8B running locally on our GPU. Stateless by design — you send context plus a query, you get an answer back:

  • Document QA — drop a long contract or report into the context, ask questions about it
  • Classification and extraction — categorise tickets, extract fields from messy text, normalise records
  • Form processing — turn free-form input into structured data
  • OpenAI-compatible token reporting — usage.prompt_tokens / completion_tokens on every response, so cost is predictable

Private RAG chatbot

Bring your own knowledge base, get a private chatbot over it. Create collections directly from the dashboard, upload documents (PDF, DOCX, TXT, Markdown), and query them immediately through the API or the chat UI. Vector search via Qdrant on our box, embeddings via bge-m3, generation via Qwen — all on the same infrastructure as transcription and the LLM API. Collections are customer-managed and segregated per account; your data stays on our box and never reaches a third party.

How pricing works

  • EUR-direct, pay-as-you-go. Top up your balance with Stripe, spend it as you use the service. No subscription, no minimums.
  • Whisper transcription: between €0.01 and €0.03 / minute, depending on the model (large-v3-turbo vs. large-v3) and the options you enable (diarisation, SRT output, translate-to-English). Clips under 30 seconds are billed at half rate.
  • Qwen LLM: €1 / 1M tokens (prompt + completion combined). Exact token counts come back on every response, so cost is predictable — a typical automation request (~5,000 tokens) costs around €0.005.
  • Auto-topup available if you’d rather not think about it: set a threshold, we charge your saved card before the balance goes empty.
  • Corporate invoicing (per top-up): we cut you a Romanian fiscal invoice (factură fiscală) for each top-up — best for clients who prefer to prepay.
  • Corporate contract (postpaid): use the service under a written contract and receive a monthly invoice that reflects your actual consumption — no advance top-ups required.

How to start

  1. Open ai.logisoftlabs.ro and create an account.
  2. Top up a balance — €10 is enough to test the full pipeline end-to-end.
  3. Issue an API key from the dashboard and start posting audio or LLM requests.

Need a custom integration, a private deployment, or a service tier we don’t list publicly? Get in touch.