@AnthropicAI — Claude can now cite its sources
Introducing Citations. Our new API feature lets Claude ground its answers in sources you provide.
Claude can then cite the specific sentences and passages that inform each response.
Smart move. The biggest practical problem with RAG pipelines right now isn't retrieval quality — it's trust. Users need to verify outputs, and without citations, that means manually hunting through source documents. Having the model return exact sentence-level citations alongside the answer cuts that verification loop significantly. The implementation detail that matters here is "specific sentences and passages" — that's much more useful than document-level attribution, which is what most teams cobble together themselves.
@tom_doerr — open-source AI text editor
Open-source text editor with AI features
Saved this mostly to check what the actual implementation looks like. The AI editor space has a lot of vaporware — tools that slap a chat sidebar onto a Monaco instance and call it done. What makes editors like Zed and Cursor interesting isn't the AI part, it's the editor architecture that makes AI integration feel native rather than bolted on. Curious whether this one goes that route or takes the simpler path.
@tom_doerr — parse PDFs, Word docs, and PowerPoints
Parse various document types like PDF, Word, PowerPoint
Document parsing is one of those problems that looks solved until you actually try to extract structured content from a real-world PDF. Libraries like pdfplumber, pypdf, and Apache Tika all exist, but they handle edge cases differently — multi-column layouts, embedded tables, scanned pages. An open-source tool that unifies PDF, .docx, and .pptx under one interface is genuinely useful for anyone building document ingestion pipelines. No fuss, no SaaS pricing tier.
@tom_doerr — open-source CRM
Open-source CRM project
CRM is a graveyard of overengineered SaaS. An open-source self-hosted CRM is one of those things that's always in demand but rarely done well — the category needs data ownership, custom fields, and decent import/export more than it needs another Kanban board. Worth keeping tabs on where this goes.
@rauchg — Tremor Labs joins Vercel
Welcome @tremorlabs to Vercel. Two of the best design engineers I've had the privilege to meet.
They'll be working on bringing their taste and UI components to @v0 with @shadcn and to the @vercel platform. Enjoy their amazing work in open source today.
Tremor builds some of the cleanest dashboard components in the React ecosystem — their chart and layout primitives are the kind of work that makes you wish more UI libraries had this level of design discipline. Plugging that into v0 and the Vercel platform makes sense: v0's biggest gap right now is that generated UIs often look technically correct but aesthetically flat. People with actual taste on the team should help.
@adamsilverman — AI Agents week in review
Here is everything that happened in AI Agents this week 🧵
(save for later)
Saved this as a reference point. The AI agents space is moving fast enough that a weekly compiled thread is more useful than following individual announcements. January 2025 was a busy week — OpenAI Operators dropped at roughly the same time, so the thread likely covers that alongside whatever else shipped. Good to have a timestamped snapshot.
@omarsar0 — OpenAI ships Operator and Agents
OpenAI Introduces Operator & Agents!
Here is everything you need to know.
The Operators announcement was the big news of the week — a structured API for giving models the ability to take actions in the real world (filling forms, booking things, navigating UIs). The interesting engineering question isn't whether this works in demos but how it handles error recovery when a web page looks different than expected. That's where most computer-use pipelines fall apart.
