Gaza~Dev @DevGaza

@thdxr You should try **Gentle-ai** or **Gentle-shell** And if you like it, I'm sure you'll be impressed. Will you give me some tokens?

Build your own harness. This is a seriously good paper from NVIDIA on self-improving agent harnesses. The paper is called SoL-Pi, out of NVIDIA, NTU and MIT. Every coding agent runs inside a harness. That's the layer that reads files, runs commands, manages memory and talks to the model. Humans tune it by hand, reading long execution logs and turning what they notice into code changes. They handed the job to an AI instead. A research AI watched a coding agent work, proposed changes to the harness, tested them, and kept only what saved tokens without hurting results. It ran about 150 research directions across roughly 500 environments, more than 3,000 runs and over 60,000 agent-environment interactions. Four fixes survived. 1. The agent edits a file and runs the test in one call instead of two. 2. It compresses its memory when a subtask finishes, and only if that's cheaper than keeping everything. 3. It sends big tool outputs in full twice, then swaps them for a 1KB excerpt with a handle to pull the rest on demand. 4. A cheap model summarizes build and test logs, and a verifier rejects the summary if it drops evidence. None of that touches the model. It's all in the software around it. On EdgeBench, a 51-task benchmark, token traffic dropped 44.7 to 49% and API cost fell by about a third against Pi, the harness it was built on. Against the native harnesses it came in 50% cheaper than Codex on GPT-5.6 Sol and 54.3% cheaper than Claude Code on Opus 5. Scores held at roughly 94% of the baseline. They built all of it on GPT-5.6 Sol, then dropped it onto Opus 5 with zero changes. It still used 44.7% fewer tokens. The cost conversation so far has been about cheaper models. This paper says a big chunk of the waste sits in the software around the model, and an AI finds it faster than we do. Their next move is running the search again with the cheaper harness, so a cheaper agent funds the search for an even cheaper one.

OSINT METHODS 39 step-by-step guides to the most common OSINT techniques. Beginner, intermediate and advanced levels. osintguide.com/en/methods #osint #socmint

「そのサービス、作る前に需要を確かめたい」人へ👀 startup-skill、⭐️約1千。事業アイデアの検証を支援するAIスキル集。 何ができるかというと👇 ・市場調査と競合の料金・機能比較 ・差別化の整理と収支予測 ・投資家向けピッチと想定問答の準備 ・見込みが薄いアイデアには厳しい指摘も 「作ったけど誰も使わない」を避けたい個人開発者に。 github.com/ferdinandobons…

HelixDB is an open-source graph-vector database built in Rust for AI memory, knowledge graphs and RAG. Instead of keeping embeddings in one database and relationships in another, Helix stores graph and vector data in the same engine, while also supporting KV, documents and relational data. Queries can combine graph traversal, filtering and vector search in one request, and the current SDKs for Rust, TypeScript, Go and Python send the same structured query AST directly to the server. It can run in memory, on disk, or against S3-compatible object storage. 6k stars · Apache 2.0 GitHub Repo: github.com/helixdb/helix-…

GLM 5.3, hosted on Mistral, is now in Vibe Code, and the default in the web app. 1M context window and generous limits included. Happy 🛳️!

mcp + codemode + sqlite developers.cloudflare.com/agents/tools/c…

tobi lutke @tobi

4 days ago

MCP vs CLI for LLMs is the wrong discussion. People are debating at the wrong layer. Both work incredibly well as long as they are run through a repl like environment that can persist state. The funny thing about CLI is that CLI tools are usually accessed through BASH which

233 174 3K 562K 3K

github.com/cloudflare/sec…

🔥牛了！微信正式开源端到端文档解析模型 WeVisDoc，一张图直接变 Markdown！ 整页图片一次输出完整 Markdown：正文走 Markdown，公式走 LaTeX，表格走 HTML。拍照合同、翻拍论文、扫描报表都能直接吃。 关键亮点： 🔹单模型端到端：版面理解、文字识别、公式识别、表格重建全包，无需流水线 🔹双版本开源：2B 轻量好部署，4B 打退化页面更稳 🔹OmniDocBench v1.6：95.38，端到端组第一 🔹PureDocBench 三赛道均值：75.54，Clean / Digital / Real 全第一 🔹公式和表格是最大优势，脏图上比很多流水线更扛造 🔹基于 Qwen3-VL-2B/4B，Apache-2.0 论文：arxiv.org/abs/2609.20423 GitHub：github.com/Tencent/WeVisD… HF：Tencent/WeVisDoc-2B / WeVisDoc-4B #WeVisDoc #开源 #文档解析 #OCR #RAG #微信 #腾讯AI​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​ #WeChat

keyboard_arrow_left Previous keyboard_arrow_right Next

SuSu_酥酥👅 @NFT_Chen

4 weeks ago

💥最新：微信不声不响把多模态检索榜打到第一，还开源了！ WeMM-Embedding是微信最会藏的多模态模型，今天自己揭开了！ 分点看亮点： 🔹文字、图片、视频、文档、图文混排，全部进同一个向量空间 🔹已落地视频号、公众号、朋友圈、电商搜索与推荐 🔹2B 就超过此前领先的 8B 开源模型 🔹9B 在 MMEB-v2

5 8 33 19K 36

A VPS costs $480 a year. This repo has given it away since 2015. Nobody pays. Not a trial. Not a credit card. Not a 14-day countdown. It's called free-for-dev. 132k stars, 1600+ people adding new free tiers and killing dead ones every single week. The rule is brutal: if it isn't free for at least 1 year, it never makes the list. If the SSL certificate isn't free too, the whole entry gets deleted. Inside: → Oracle Cloud - 2 servers, 12GB RAM, free forever, no time limit → AWS Lambda - 1 million runs a month before you pay a cent → Cloudflare - unlimited sites, DNS, SSL and a global CDN → Google Cloud Run - 2 million requests a month on the house Most providers: $40 a month for 1 server. Oracle: 2 servers for $0. Every side project starts with the same 20 minutes of Googling "is this actually free." This repo did that search in 2015 and has redone it every week for 11 years. Bookmark it before your next project makes you search all over again. repo below

SECTOR @sector_sol

4 weeks ago

x.com/i/article/2093…

1 1 24 104K 51

@rileybrown Memory: ENGRAM Harnes: Gentle-ai The two tools created by the master: @G_Programming

Firefox hides dozens of powerful settings. Most people never touch them. A few changes in about:config can reduce tracking, stop hidden preloads, and make Firefox feel faster without breaking your favorite websites. Here are the ones worth changing:

@G_Programming Hay que esperar a que salga un modelo parecido OpenSource que seguro acabara apareciendo

Cloudflare ha lanzado su propia skill completísima de auditorías de seguridad. Es la que utilizaban internamente en la empresa como sistema de detección de vulnerabilidades. → github.com/cloudflare/sec…

卧槽， 终于排上了kaggle TPU，我把GLM-5.3-Flash 跑起来了， 它在 hashneuron.space 上运行的飞快！

王小庄 @BlockInsight214

a week ago

使用 kaggle 免费GPU，不用排队一键运行推理模型 DeepSeek，每周30小时额度。 kaggle.com/code/crossover…

7 18 146 31K 247

@minchoi There is nothing better currently than the "Gentleman Engram" memorial

@thdxr They should add it; it would bring about very interesting things for both users and opencode.

Zvec v0.7.0 is here! 🎉 This release expands the Zvec ecosystem and brings major upgrades to indexing, deployment, and usability. Upgrade today and give it a try! ✨ What’s new 🧭 zvec-grep (github.com/zvec-ai/zvec-g…) A local-first workspace search CLI for humans and AI agents. It combines semantic search, BM25, and rg to find code and documents faster—with fewer search rounds, tokens, and less time spent. 🔗 ReMe integration Zvec can now serve as a file-store backend for ReMe, providing in-process HNSW ANN search for agent memory management. 🚀 Production-ready DiskANN Now supports Linux ARM64, macOS ARM64, and io_uring asynchronous I/O—with automatic fallback to the best available backend. ⚡ Upgraded index algorithms Introduces the new IVF-RaBitQ index. RaBitQ now supports runtime AVX2/AVX512 dispatch, automatically selecting the optimal path for each CPU. 📦 Better deployment experience macOS ARM64 C API library reduced from 37 MB to 22 MB—40% smaller New musl libc and Alpine Linux support Prebuilt SDK binaries for Linux, macOS, Windows, Android, and iOS with every release 🛠️ Improved usability New DocIterator for streaming full-collection document traversal New N-gram tokenizer for phrase, code, and short-text search 📚 Learn more zvec.org/en/blog/2026-0… We’d love to hear your feedback! 🚀

We’re launching the Cline Ambassador Program. If you’re a developer or technical creator who enjoys building with Cline and sharing what you learn through projects, blogs, videos, or social content, we’d love to hear from you. As an ambassador, you’ll get early access to what we’re building, exclusive perks, and opportunities to work more closely with the Cline team. The top ambassadors will be promoted to paid partners. Apply now: forms.gle/3ZoHtwtDmyNqAo…

same method, new list of models: > cl/deepseek/deepseek-v4.1-flash > cl/meta/muse-spark-1.3-contributor > cl/z-ai/glm-5.3-flash > cl/stealth/union-alpha if there is nothing change, same number: > 15M tokens/ model/ account/ daily > totally FREE, in your favorite harness you do the math,

Luong NGUYEN @luongnv89

2 weeks ago

x.com/i/article/2097…

5 11 99 69K 239