Raphael Mansuy 🍵

Truth is Expensive: Why RAG is Doomed and What Actually Matters You're not building a truth machine. You're building the world's most efficient fiction amplifier. That's not a bug. It's the inevitable result of solving the wrong problem. Yuval Noah Harari's core insight—delivered in his latest book Nexus and his viral TED thesis—is brutally simple: Humans are a post-truth species. We didn't conquer the planet by being right. We conquered it by believing in shared fictions: gods, nations, money, brands. These stories don't need to be true. They need to be believed. [1] And now we're hardcoding that vulnerability into AI.

The more I play around with GEPA the more I realize its not just a prompt optimization algorithm. Its really the most efficient way to have an LLM explore a massive dataset and and make useful insights. This visualizer alone shows just how cool that process is.

10 GITHUB REPOS THAT WILL CHANGE HOW YOU USE THE INTERNET. Bookmark every single one. Most people have no idea these exist. 1. LocalSend / github.com/localsend/loca… AirDrop for every device. Share files between Windows, Mac, Linux, Android, and iOS with no account and no cloud. 2. FreeTube / github.com/FreeTubeApp/Fr… Desktop YouTube player with zero ads, zero tracking, and zero Google account required. 3. Vaultwarden / github.com/dani-garcia/va… Self-hosted Bitwarden server. Replaces 1Password, LastPass, and Dashlane for $0 forever. 4. Syncthing / github.com/syncthing/sync… Replaces Dropbox, iCloud, and OneDrive. Syncs files directly between your devices with end-to-end encryption. 5. Ladder / github.com/everywall/ladd… Bypasses paywalls on NYT, WSJ, Bloomberg, Nature, and hundreds of other sites. Runs on your own server. 6. Immich / github.com/immich-app/imm… Self-hosted Google Photos replacement. Auto-backup, facial recognition, and AI search on your own machine. 7. AdGuard Home / github.com/AdguardTeam/Ad… Network-wide ad blocker. Blocks ads and trackers on every device in your home including smart TVs. 8. Jellyfin / github.com/jellyfin/jelly… Self-hosted Netflix. Stream your own movies, shows, and music to any device with zero subscription. 9. LibreChat / github.com/danny-avila/Li… Run ChatGPT, Claude, Gemini, and DeepSeek in one interface. Pay only for API usage. No $20/month subscription. 10. Uptime Kuma / github.com/louislam/uptim… Beautiful self-hosted monitoring for every website, server, and service you own. Replaces $50/month tools. Big tech built the internet you use every day. These repos built the one you actually want.

This is HOW my Hermes and OpenClaw Agents share a single brain. Hermes Agent is the orchestrator with OpenClaw agents as the squad.

Kafka, but for agents!

Anyone tried in Hermes yet? New OS king?

For those not taking full advantage yet of the new Hermes agent / X 0Auth integration here's an info graphic and fun way to learn about this new cool integration! what do you like most about this new cool feature?

If you can code but can’t figure out what to build there’s tons of alpha in here Start building. x.com/indutripat8242…

@victor207755822 Nice !

Bty，the owner of this project is one of our most gifted （richest, also🤣）workmate: Tianyi(linkedin.com/in/tianyicui/), don't hesitate to join us~

Oh my god it scored worse than Composer 2! Not even 2.5! And it cost 4x more to run!!! This might be the worst major lab model drop of all time. Llama 4 tier. Insane.

@DFintelligence No taxes at all will be better😎

Je rêve d’avoir une API impots.gouv simple, et de demander à Codex de se connecter à mes justificatifs, ma boîte e-mail, mon compte bancaire, et de faire toute ma déclaration d’impôts automatiquement, puis d'organiser tout les justificatifs comptable proprement dans un dossier. Si on arrive à mettre toute l'administration en API purée mais ca serait incroyable.

Private equity secondaries is a multi-billion dollar market still run on PDFs, emails, and relationships. We're replacing that with AI-native infrastructure — a multi-agent platform where intelligent agents aren't a feature, they're the foundation. I'm hiring two founding engineers for our AI team: - Senior AI Engineer — You've shipped production-grade AI systems. You can architect a platform, mentor a team, and still write code that makes people stop and look. - Junior AI Engineer — You're obsessively curious, self-taught, and building toward becoming one of the best AI engineers in the world. You just need the right environment. You'll work directly with me. Real autonomy. Modern stack (Anthropic, React 19, AWS, Terraform). The scoreboard: real capital flows, not dashboards nobody checks. Must already be a permanent resident or have right to work in Hong Kong or Shenzhen. No visa sponsorship. No recruiters. No portals. CV + a short note on the most interesting AI system you've built. Send directly. (Details in first comment.) --- 廣東話版 私募股權二級市場係一個數十億美元嘅市場，但到依家仲係靠 PDF、電郵同人脈關係運作。我哋要用 AI 原生基礎設施取代呢一切 —— 一個以智能代理為根基嘅多代理平台。 我而家搵兩位創始工程師加入 AI 團隊： 資深 AI 工程師 —— 交付過生產級別 AI 系統，識設計架構、帶人，代碼令人眼前一亮。 初級 AI 工程師 —— 對技術有近乎癡迷嘅好奇心，自學能力極強，目標係世界頂尖。你只需要一個啱嘅環境。 直接同我合作，真正嘅自主權。技術棧：Anthropic、React 19、AWS、Terraform。衡量標準：真實資本流動。 必須已持有香港或深圳嘅永久居留權或工作權利，唔提供簽證擔保。 冇獵頭、冇申請表。履歷 + 你做過最有趣嘅 AI 系統簡介，直接 send 過嚟。（詳情見第一則留言。） Send me a PM

I just built a Multimodal Agentic RAG using Gemini Embedding 2 and Google ADK. Any input. Cited answers. Every source visualized in a live 3D embedding space. 100% Opensource.

dflash-mlx sweet spot on Apple Silicon is narrower than expected: prompt ≤16K AND max_tokens ≤1024 to get the 1.45-1.65x decode speedup. the speculative win is concentrated in the first ~500 generated tokens, then acceptance decays and the gap closes. outside that, llama.cpp wins on wall time. dflash for short prompt + short reply (one-shot, data gen, classification). benched M3 Ultra 512GB, Qwen3.6-27B-Instruct-8bit + z-lab DFlash draft vs llama.cpp build 9010. 5 prompt sizes × 3 max_tokens. where dflash hurts (cold vs cold): - prefill 2-3x slower. 32K prompt: ~220 vs 676 tok/s - end-to-end at 16K + 300 tok reply: 66.8s vs 46.3s. 32K: 2.5x slower - long gen (2048 tok): tied or llama wins regardless of prompt size

Super excited that @cocoindex_io made to Python trending! Built on Rust Core, CocoIndex provides ultra performant python-native experience that turns codebases, PDFs, Slack, inboxes, and videos into a continuously updated index your LLM app can reason over. CocoIndex has been showing up on Rust trending quite a few times in the past year. and with V1 launch, we have added more connectors and examples and made python takes bigger half of the repo. Stay tuned as we ship Rust SDK and more examples, and we may go back to Rust trending again later ;)

Google Agent Development Kit is super underrated. I will look at: - Agents CLI - Ambient and Resume Agents - Context Management - Long term Memory - Agent Evals (User and env simulation)

This open source project is getting viral on linkedin ! - Your agents deserve fresh context. @cocoindex_io turns codebases, PDFs, Slack, inboxes, and videos into a continuously updated index your LLM app can actually reason over … no stale nightly batches, no context gap. Python on a Rust delta engine: edit a chunker or swap embedding models, only the affected rows re-run. Pgvector, LanceDB, Neo4j as targets. Byte-level lineage per row. Production agent ready in 10 minutes. Link to OP in comments!

I just built a Voice AI Agent for insurance claims using Gemini 3.1 Flash Live and Google ADK. Talk to it. It fills the intake form, extracts claim details, and routes to an adjuster in real-time. 100% Opensource.

Advice for AI engineers 💡 Browser control is possible with a Small Model, like LFM2-350M by @liquidai . Here's a 60-minute deep dive on how to fine-tune with RL and OpenEnv by @huggingface Enjoy ↓