Open-Weight vs. Open-Source AI: What’s the Difference?

A few weeks ago we posted the open-source Graphsignal profiler here: wrap vllm serve / sglang serve / whatever in graphsignal-run and it serves everything it measures as JSON, mainly so an agent can read it in a loop. Turns out we kept opening the JSON ourselves to see what the GPU was doing. So now there is graphsignal-top : graphsignal-run vllm serve <model> --port 8000 # one terminal graphsignal-top # another What you get: per-GPU util over time, memory, power, throttling; the engine's running/waiting requests, tok/s and TTFT/TPOT p50/p95; a GPU time split (kernels / graphs / memcpy / sync); and every kernel ranked by time over the last 5 s with a readable name and category, not just the mangled symbol. Tabs for CUDA graph replays, transfers and syncs. Screenshot is Qwen3-32B on 2x RTX

Hey guys, I built LetsFG with a friend of mine and our open source community. It's an MCP that lets Claude search and book hotels and flights. Today we got listed live in Claude's official connector directory. When we started I couldn't find much about what the review actually looks at, so here's everything we had to change, in case you're building a connector too. Tool descriptions must describe, not instruct Our book_flight description was like 5,000+ characters of "always do this, never do that", basically steering the model through the description, and that's not allowed. A description just says what the tool does, what it takes and what it returns. So we moved all the "what to do next" stuff into the tool results, and honestly it works better there coz Claude reads it at the exact mom

I wanted Codex to do things on my phone, not just in a terminal, so I built an Android app for it. How it works: the app carries a small Linux system (Alpine under proot, no root) and runs the real Codex CLI inside it, signed in with your ChatGPT plan, with no API key. The phone is given to Codex as tools: read the screen, tap, type, scroll, open apps. Codex is already good at tool loops, so it just works through the apps like a person would. What I use it for: turning on my lights, ordering food and changing the order while it works, replying to messages. It can also use ChatGPT's voice for dictation and reading replies aloud. Before anything that sends, posts or buys, it shows an approval card with Approve, Deny or Change. It's free and open source (Apache 2.0), and works with Claude Cod