Laurent Wiesel 🇮🇱⭕🤖

@jerryjliu0 @llama_index Og

Our core mission today is using AI to solve document OCR. All of our product offerings, from commercial (LlamaParse) to open-source (LiteParse, ParseBench), are fully aligned towards solving this problem. Introducing our revamped website 👇 llamaindex.ai/?utm_medium=so…

@jerryjliu0 I agree

LiteParse is the best model-free, open-source document parser for AI agents. It now gets a first-class landing page on our website 💫 Our company mission is building the world's best agentic document processing platform, and liteparse is the central pillar behind our OSS efforts. It's blazing fast (and getting faster soon!), supports 50+ file formats, and is one-shot installable as an agent skill. Webpage: llamaindex.ai/liteparse?utm_… Come check it out: github.com/run-llama/lite…

@jerryjliu0 Love

I built a Claude Code skill that allows it to generate a deep research report over any collection of complex docs (PDFs, Word, Pptx)….and generate word-level citations and bounding boxes directly back to the source! 📝 Check out “/research-docs”. 1. It parses out text and bounding boxes from every doc with liteparse, in seconds. 2. It then generates a full HTML report of the outputs that let you see word-level citations in each page. Raw Claude obviously has deep research capabilities, but it lacks an audit trail back to the source. This skill gives you a researched report that can be audited by others. Check it out: github.com/jerryjliu/lite… LiteParse: github.com/run-llama/lite…

@bcherny @jxnlco What about pi.dev? Can we continue using oauth with Pi?

Starting tomorrow at 12pm PT, Claude subscriptions will no longer cover usage on third-party tools like OpenClaw. You can still use these tools with your Claude login via extra usage bundles (now available at a discount), or with a Claude API key.

@cathrynlavery Thanks looks great.

🦞 openclaw-ops tired of babysitting your OpenClaw? I just open-sourced my ops layer. It fixes the gateway, exec approvals, broken crons, stuck sessions, channel issues, and security gaps. • heal.sh — one-shot fix for the most common gateway issues (auth, exec approvals, crons, stuck sessions) • watchdog.sh — runs every 5 min, restarts gateway if down, escalates after 3 failures • watchdog-install.sh — installs the watchdog as a macOS LaunchAgent so it survives reboots • check-update.sh — detects version changes, explains what config broke and why; --fix to auto-apply • health-check.sh — declarative URL/process checks for gateway-adjacent services and workers • security-scan.sh — config hardening score (0–100), drift detection, credential scan • skill-audit.sh — static audit for third-party ClawHub skills before you install them basically everything I built for myself since January to stop me from tearing my hair out 💀 link below 🫶

@rcarmo Lol

"Feature creep has been promoted from lifestyle choice to corporate strategy." is our new tagline.

People of pi, github.com/rcarmo/piclaw/… is now a full-blown productivity suite, because somewhere last week I took a wrong turn into Windows land and I just can't stop. I promise I am getting back to squads and agents soon. /cc @badlogicgames

@mehdirhasan Lol dumb ass

Google 1967, dumbass

Do you remember when you joined X? I do! #MyXAnniversary

This is a great video summary of what @Chabad stands for, and a powerful response to Tucker, Candace, and all the haters out there. We are Chabad.

so you're telling me i can now... embed a video embed a voice memo embed a PDF embed an image embed text ...all in the same space? with one model? and search across all of them with a single query? time to rebuild everything.

This is an appreciation post for the hard work the Chabad Emissaries do all around the word for so many peopl. Give a shoutout to the Chabad rabbi and rebbetzin in your life!

@ArielSterman Shabbat shalom

Parshat Ki Tisa. You shall carry. The message this week for Jews around the world is clear. Stay safe and eyes open!

I ♥️ chabad. Shabbat shalom motherTuckers.

Some thoughts. Not sure if anything actually changed but today felt like a genuine escalation in anti-Semitic rhetoric that we haven’t see in the mainstream for some time. Blaming Jews for starting wars was the battle cry of Nazi Germany. Hitler explicitly justified the Holocaust bc of this libel. As absurd and even hilarious as the antisemites might be it is also seriously frightening. Society has very little room left to save itself before truly descending into some unthinkably dark places. As I’ve repeated: The purpose of Judaism is not to fight antisemitism, we fight antisemitism in order to focus on the purpose of Judaism. I know some really bad people posed the q, but that’s not an excuse: we need to have answers. So, what is the Jewish vision for the State of Israel? What is our vision for the future of Judaism? These are genuinely important questions—not for antisemites but for ourselves. Now is the time to really think and consider what is your vision of redemption? What is your vision for a Jewish State? We need to have the answer for these questions or, God forbid, our enemies will. The Jewish People will survive. נצח ישראל לא ישקר Not everyone will be able to stop antisemites or influence global politics but there is a step everyone can take to bring themselves a little closer to the Jewish People and God’s unbreakable covenant. As the Rebbe said, “if you know aleph, teach aleph.” Start living the Judaism of your dreams right now. עם ישראל חי

Anybody noticing something up with Claude Code? Or is it me?

@jpschroeder I'm with you brother. Keep shipping!

dmux 5.4.0 is out - QoL update. More control and faster workflows: - Set min/max pane widths to your preference - Add a non-agent shell pane to any active worktree - Auto dmux control pane restore if crashed / closed. - Improved UX for agent selection dmux.ai

Something is wrong with Claude Code today.

We're watching a three-stage evolution in how AI agents manage context. Stage 1: Tool bloat. Give the agent every tool, every connector, every context; tool descriptions, retrieved docs, conversation history. Performance degrades. Most production agents are still here. Stage 2: Sub-agent delegation. The orchestrator delegates to specialist agents who each handle their task and return compressed results. Better, but the orchestrator's context window is still the bottleneck for the user's input. If someone hands you a 10M token codebase, no amount of sub-agent delegation fixes the fact that the input itself doesn't fit. Stage 3: Prompt as external object. Here, the prompt isn't something the model reads; it's something the model navigates. The data remains external, and the model writes code to decide what to pull in. This is moving fast. MIT's Recursive Language Models paper dropped in January. By February, major labs are shipping production versions of the core idea. Instead of feeding a 10M token prompt directly into the model, the RLM loads it as a Python variable in a REPL environment. The model never sees the full content, it only gets metadata. From there, it writes Python code to slice into specific sections, run regex searches, filter what's relevant, and call itself (or a smaller LLM) on each chunk. The goal is not to dump raw context into the window, but to write code to filter it before it enters the window. You spend a little compute on filtering, get better quality and lower cost because you're not filling the context with irrelevant stuff. Today it's "filter search results with code before they hit context." Tomorrow it's "the 500-page patient chart lives as a variable, and the model writes regex and sub-calls to navigate it." We're moving toward systems where the model automatically handles context engineering during inference, programmatically.

Darnit

@emollick I don't think so, but have considered morphing your imagegen agent into a Skill?

Does Claude Cowork support agents? I know it launches its own subagents, but can you import agents from Claude Code into Cowork? For example, I have an Code image creation agent that uses the GPT-imagegen API, I want to do image generation in Cowork, does anyone know how?