Gladia

What if every call flagged its own quality issues automatically? No manual sampling. Every call scored as soon as it ends. That's what Gladia's voice pipeline makes possible for contact center and QA/CX teams. Here's what teams are building with it: • Live calls: real-time compliance alerts at • Full-context calls: async batch scoring for QA and trend analysis • Every transcript: sentiment, talk/hold/silence time, and overtalk detection • Flagged interactions: speaker-attributed records routed to coaching • All calls: structured payloads pushed to CRM, WFM, or QA platforms One API call turns raw audio into structured and ready-to-use data. You choose where it goes. Full breakdown in comments.

Your coding agent now knows Gladia by 💜 Because we just shipped Gladia Skills — official context files that give Cursor, Claude and other AI coding agents built-in knowledge of the Gladia API and SDK. That means no more hallucinated parameters or digging through docs mid-build. Six skills cover the full path: • SDK setup • Pre-recorded transcription • Real-time streaming • Audio intelligence (diarization, translation, PII redaction, subtitles, and more) • CLI transcription • Troubleshooting Install in one line → npx skills add gladiaio/skills. Try it, break it, or send a PR

✨ You know an evening was magical when it ends with over a hundred check-ins, caviar, nuggets, a FIFA lego raffle, and a room that never went quiet. That was our meet-up last week. Thank you to everyone who came, and to Vercel, Modal, Poolside, turbopuffer for hosting it with us. A few frames below

Your terminal just learned a new trick — Gladia CLI, transcription in one command! Built for developers who want to skip the dashboard and go straight from audio to transcript. What you get: • One command to transcribe • Flexible output: plain text, JSON, SRT, or VTT • Best-in-class speaker diarization powered by pyannoteAI Precision-2 • Multilingual support with 100+ languages • Choice of models: solaria-1 or solaria-3 Install it, set your API key, and you're transcribing in under a minute. P.S. We tested it on a fun one: Mbappé correcting an interviewer on his World Cup record, transcribed end-to-end

Big news for healthcare enterprises: Gladia now supports HDS hosting — one of the strictest health data security frameworks in Europe. HDS is France's mandatory standard for any platform processing personal patient data. For a long time, it was the compliance blocker keeping voice AI out of clinical workflows. Not anymore. Real-time transcription, medical dictation, and clinical call centers are all now compliant with the French health data law. If you're an enterprise building in healthtech, the door is open.

We're heading to RAISE Summit 2026! 🇫🇷 • July 8–9 • Booth S5, Le Carrousel du Louvre, Paris Thousands of AI builders, researchers, and decision-makers in one place, right next to the Louvre. Not a bad spot to talk about the future of voice AI, huh? Come find us at S5. We'll show you what we've been shipping, geek out on transcription, or offer to grab a coffee in between sessions with Jeremy Lauglaney and Lisa Surraco We’re also hosting a tech meet-up on July 9th, 7–10 PM at Gladia/Poolside office if you’re in for champagne, caviar, and great company. Register below. See you in Paris 👋 #RAISESummit2026 #VoiceAI

Quality monitoring used to be one of the most time-consuming jobs in a contact center. Most teams can only review 5% of interactions manually. The rest goes unused. Gravite.io built an AI-powered quality monitoring platform to automate this process. But making that promise real meant finding a transcription API that could handle noisy, real-world French telephony audio, at thousands of hours a month — and they found it in Gladia. The result: complete call coverage, and review time cut by 93%.

🥂 Next week, we're hosting a tech meet-up together with our friends at Vercel, Modal, Poolside, and turbopuffer. If you're a founder, engineer, researcher, or model builder, this one's for you. On the menu: champagne, caviar, nuggets, and genuinely good company in one of the best summer settings Paris has to offer. Spots are limited, and attendance is curated. If you're around for RAISE Summit or just happen to be in town, register through the link in the comments. 📆 Thursday, July 9, 7–10 PM 📍 Gladia/Poolside office, rue Sainte-Cécile, Paris Anna Jelezovskaia, Cecilia Caron, Arina Sugako, Will Kenney, Tracy Mendoza, Evelyne Couprie, Andrea Badia Serra, Constant Dacbert Maxime Gaudin

Last week we found out our open STT leaderboard had been RIGGED. Not even kidding. A few months ago we built compare-stt — a blind platform where anyone can upload audio, get transcriptions from two anonymous providers, and vote for the better one. Elo-ranked. We gave it to the industry for free because we were tired of benchmarks that could be optimized for. We wanted the users to judge for themselves. No brand bias, just transcript quality as perceived by the naked eye. It blew up quite a lot, and we knew that everyone in the industry who mattered had seen it. But someone found a way to cheat! A few weeks back, we spotted a strange pattern: one provider kept climbing back to the same position within hours of any ranking shift. Too fast. Too consistent. How? Vote tokens weren't single-use. Anyone could script a loop - upload, transcribe, vote - and inflate Elo as fast as their connection allowed, and that’s what they did. We pulled the logs, patched the holes, ran forensic cleanup. Published everything, incl. the vulnerability writeup, the fix, the cleanup script, on GitHub. Then we recomputed. The ranking went back to what it was before the weird stuff happened. And as of today again, in all honesty and transparency, Gladia holds the #1 spot. I guess, when people game your benchmark, they're telling you the benchmark matters. I find that genuinely touching. But guys, let’s keep this honest. And no, we won’t say who it is - not our DNA. But wink, wink we know who you are.

Gladia now integrates with Attendee, an open-source Meeting Bot API. Drop a bot into any Zoom, Google Meet, or Microsoft Teams call with one API call. Get real-time transcription, automatic language detection, and per-participant speaker diarization on the final recording. No RTC plumbing required. What you can build from day one: • Meeting assistants with notes, summaries, and action items • LLM agents that join calls and understand speakers in real time Full integration guide in the comments.