LightOn

🔥 FastPlaid 1.10.0 is live

now you can update your index with new documents without starting from scratch.

Driven by @raphaelsty.bsky.social, FastPlaid makes multi-vector indexes easy, fast, scalable, and perfect for batch workflows. 👏🏻

👉 github.com/lightonai/fa...

#FastPlaid #VectorSearch

If your AI lives in someone else’s cloud, it’s not really yours.

Bring it home with LightOn.

🛡️ t.co/1bg98xs1kA

From soul-sucking RFPs to game-changing proposals
LightOn’s GenAI strips out the 90% noise,finds the real opportunity,and writes the answer that shows how you’ll change your customer’s life,fast, precise,and backed by every doc that matter
No more blank-page dread,Just hit send and let your idea win

After reaching maturity, the PyLate library has its paper accepted at #CIKM2025
🎉Congratulations to the two creators of the library @nohtow.bsky.social @raphaelsty.bsky.social

📄 Read the paper: lnkd.in/eh6NQj4B
Discover PyLate : lightonai.github.io/pylate/
🌐 Learn more about LightOn: lighton.ai

Most GenAI reads documents like a headline.
It skips tables, charts, footnotes, and logic.
LightOn reads them like you 🫵🏻

🚀 Deploy your own AI → www.lighton.ai

La solution ? ➡️ Inverser la logique
Au lieu d'envoyer vos données vers une IA, faites venir l'IA directement sur vos infrastructures.

Pour en savoir plus sur la vision de LightOn 👉https://www.maddyness.com/2025/07/17/ia-generative-vous-avez-dit-souverain/

🔐 Privacy : protéger vos données sensibles

🔗 Indépendance technologique : réduire les dépendances externes

⚙️ Résilience opérationnelle : garantir la continuité de service

(2/3)

IA générative : vous avez dit souverain ? La nervosité des marchés sur les valeurs technologiques vient nous rappeler l’interdépendance de la technologie.

"Le contexte géopolitique actuel remet en question la pérennité de l'accès mondial aux grandes plateformes d'IA générative"

@ldaudet.bsky.social DG délégué et co-fondateur de LightOn, soulève les principaux enjeux de la souveraineté numérique (1/3)

www.maddyness.com/2025/07/17/i...

👉 To know more about ModernBERT: www.lighton.ai/lighton-blog...

📍 ACL 2025: Encoders-only coffee chat anyone?

@nohtow.bsky.social & Oskar Hallström are in Vienna to present ModernBERT Paper at @aclmeeting.bsky.social

📅 Don’t miss the Poster Session today 11am.
☕ Or feel free to catch them in the #ACL2025NLP aisles!

Congrats, your GenAI output impressed everyone! 🤝
Except your boss. And your security team...

LightOn : GenAI, on your terms.
👉🏻 www.lighton.ai

Fiers d’être partenaire des Trophées de l’innovation 2025,
Organisés par @LBP AM, qui récompensent les entreprises innovantes et engagées.

Vous êtes une jeune société innovante avec un modèle économique durable?

⏰ Candidatez avant le 01/09/25
👉🏻 Infos&candidature www.lbpam.com/fr/notre-ide...

What if you could actually Know Your Customer in minutes?
And spend more time welcoming them!

With LightOn's private AI, automate your KYC, while staying fully compliant.

👉 www.lighton.ai

Agentic AI vs. Rules-Based Automation: How AI Is Reshaping Finance - LightOn Discover why legacy automation falls short in finance, and how agentic AI brings context, adaptability, and secure intelligence to the frontlines.

Finance is shifting from static rules to strategic initiative.

AI spots patterns, flags risks, drafts decisions, without being told.

Discover where deterministic systems shine, agentic AI takes over, & how combining both drives real-time automation.

👉 Read more: www.lighton.ai/lighton-blog...

Introducing Ettin Suite: the SoTA open recipe to outperform existing Generative & Retrieval Models - LightOn Introducing Ettin, the first ever SOTA suite of paired encoder & decoder models, developed by Johns Hopkins University in collaboration with LightOn.

Kudos @orionweller.bsky.social and @nohtow.bsky.social for this tremendous work!

Check out the full details 👉 lighton.ai/lighton-blog...

The result?
🦾 Models that outperforms SoTA models at their sizes, including Llama 3.2 1B.

Ettin sets a new state-of-the-art for open-data models in both categories and opens the door to further explore whether a competitive encoder can be built from a decoder, and vice versa.

The revolution started by ModernBERT for encoders is now reshaping the entire Transformers landscape. Its breakthrough training recipe and architecture, plus a new open-source dataset, has now been applied to decoder-only models.

✊ Transformers... Assemble!

Introducing ♊ Ettin Suite, a SoTA open recipe to outperform existing Generative & Retrieval Models.

Developed by Johns Hopkins University in collaboration with LightOn, Ettin is the first-ever SoTA suite of paired encoder & decoder models.

Going Further: Late Interaction Beats Single Vector Limits Single vector search is the standard for RAG pipelines, but struggles in real-world applications due to poor out-of-domain generalization and long-context handling. Multi-vector models overcome these ...

📍Listen now: maven.com/p/1973fe/goi...

Want to learn more about sovereign Search AI and how LightOn is leading the way? 👉 lighton.ai

Going Further: Late Interaction Beats Single Vector Limits Single vector search is the standard for RAG pipelines, but struggles in real-world applications due to poor out-of-domain generalization and long-context handling. Multi-vector models overcome these ...

Forget single-vector limits. This is the new wave.

🍽️ On the menu: out-of-domain, long-context and reasoning-intensive retrieval, and hands-on tooling with PyLate

🧠 Tune in to see what Chef Antoine’s been cooking.

🎧The future of generative AI is context-rich, data-deep, and finally able to read at scale.

@nohtow.bsky.social from LightOn joined Hamel Husain and Shreya Shankar to break down how multi-vector search is rewriting the rules of knowledge management

🔍 From Document Overload to Instant Insight
When most AI drowns in data, LightOn swims through millions.

👉 Ask your question, it hunts the answers.
👉 No fragments. Just full-picture clarity.

No more clicking through folders or second-guessing sources.

You ask, it knows.

GitHub - lightonai/pylate-rs: PyLate efficient inference engine PyLate efficient inference engine. Contribute to lightonai/pylate-rs development by creating an account on GitHub.

💫 And here’s the magic: PyLate-rs compiles Rust code to WebAssembly, letting you run late-interaction models right inside your browser to build awesome demos!

Start compute some embeddings now: github.com/lightonai/py...

👏 Kudos Raphaël for this new milestone!

🏎️ Introducing PyLate-rs

@raphaelsty.bsky.social is back at it to make you love late-interaction models!

After quickening the retrieval process with FastPlaid, he’s now making inference lightning-fast with this lightweight tool crafted in Rust for optimal speed and efficiency!

AI is already running in your company, the good news is that you can secure it!

LightOn puts Generative AI where it belongs: inside your walls, under your watch.

🚀 Deploy your own AI: lighton.ai

🚀 LightOn is accelerating fast across AI Research, Engineering, Product & Go-To-Market.

Come contribute to one of Europe’s leading players, in the most thrilling industry of the moment: Generative AI

👉 lighton.ai/careers

🔍 Pour en savoir plus sur l’IA souveraine de Lighton : lighton.ai

📰 Lire l’article des Echos : www.lesechos.fr/idees-debats...

🌬️ Fuite des données et du savoir-faire
🇪🇺 Perte de souveraineté technologique

”Les employés ont souvent recours à ces pratiques par nécessité” poursuit Laurent, invitant les entreprises à sécuriser le travail de leur employés avec une solution souveraine apportant performance et sécurité.

"L’usage non encadré des IA est une faille silencieuse pour les entreprises, mais pas une fatalité"

Dans un article des Echos sur le phénomène #ShadowAI, @ldaudet.bsky.social revient sur le déploiement de l’IA en entreprise et le coût d’un usage non maitrisé :

YouTube video by LightOn LightOn – Turn Your Enterprise Knowledge into Intelligence

🎥 Watch it in action: www.youtube.com/watch?v=GMtiTJB61no