DevHunt Daily
Sunday, November 15, 2026
- 1AIHacker News
DeepSeek V3 implemented in ~200 lines of pure JAX
A GitHub repo provides a compact JAX implementation of the DeepSeek V3 model, covering model definition, training, sampling, data preparation, and tests in roughly 200 lines. Developers can use it to experiment with the open-weight DeepSeek model without extensive codebases.
- 2AIHacker News
Serving GLM-5.3 across two NVIDIA DGX stations with vLLM
NVIDIA demonstrates how to run the 753‑billion‑parameter GLM-5.3 model on two DGX stations using tensor parallelism and vLLM, achieving up to 607 tokens per second. Developers can follow the guide to deploy large models that exceed a single GPU's memory.
- 3SecurityHacker News
Project Zero shifts to a 90‑day disclosure timeline, plans 60‑day then 30‑day limits
Google's Project Zero announced it will disclose reported vulnerabilities after 90 days, moving to 60 days in early 2027 and 30 days in 2028. The change aims to speed up vendor remediation and reduce the gap between patch release and user installation as AI accelerates exploit development.
- 4Dev toolsGitHub Trending
Alibaba releases OpenCodeReview, an AI-powered CLI for precise code reviews
OpenCodeReview is a command‑line tool that uses LLM agents to analyze Git diffs and add line‑level comments, supporting multiple languages and custom rules. It claims higher precision and lower token usage than generic agents, aiming to reduce false positives in CI pipelines.
github.com893 stars today - 5Dev toolsGitHub Trending
Uptime Kuma offers self‑hosted monitoring with Docker and native installs
Uptime Kuma is an open‑source, self‑hosted tool for monitoring website, service and container uptime. It provides a web UI, multiple notification channels and supports Docker, Docker‑Compose and manual Node.js installations, which is useful for developers needing on‑premise health checks.
github.com430 stars today - 6AIGitHub Changelog
GitHub Copilot app adds general availability for GitHub Enterprise Server support
The Copilot app (v1.2.0+) now lets users add a GitHub Enterprise Server account to work with remote repositories, issues, and pull requests. Cloud sessions remain unavailable for GHES. Developers using self‑hosted GitHub can now access Copilot features directly from the app.
github.blog - 7CloudVercel
Vercel Daemons can now send OpenTelemetry traces to Vercel Tracing
Vercel added support for daemons to export OpenTelemetry traces via a Unix socket, using @vercel/otel 2.3.0+. Developers can view daemon spans in the Vercel CLI, enabling tracing of background workers alongside other project traces.
vercel.com - 8AIVercel
Vercel Token Pass raises monthly token credit to $150
Vercel’s Token Pass add‑on now provides $150 of AI Gateway tokens each month, up from $80, while keeping the price at $30 per team. The credit covers open‑weight models like Kimi, GLM, and DeepSeek, and resets each billing cycle with no rollover. Developers can enable it via the Vercel CLI or API key.
vercel.com - 9SecurityCloudflare
Cloudflare warns of rapid rise in AI-driven penetration testing frameworks
Cloudflare reports a surge in open‑source LLM‑based pentesting tools like ARTEX and its many forks, with thousands of repositories and a 30× increase in daily requests. The firm advises adopting a Modern Security Model with WAF, bot defense, identity‑gated access and threat intelligence to mitigate these autonomous attacks.
blog.cloudflare.com - 10DataSimon Willison
datasette-llm adds OpenAI‑compatible endpoint support
Version 0.1a11 of the datasette-llm plugin adds support for OpenAI‑compatible endpoints, allowing any LLM that follows the OpenAI web protocol to be configured with Datasette. This enables developers to run local models from LM Studio, Ollama or llama.cpp with Datasette agents via uvx.
simonwillison.net - 11AIProduct Hunt
CodeMesh reduces AI coding token usage by up to 95%
CodeMesh creates a live map of a codebase and supplies AI coding agents only the relevant parts, cutting token consumption and API costs dramatically. In tests on a real repository it lowered token use by 95% and costs by over 90%, benefiting developers using Claude Code, Cursor or Windsurf.
producthunt.com - 12SecurityProduct Hunt
Antelog launches to monitor and block risky AI coding agent actions
Antelog, a free tool, records every command, website visit, and data transfer made by AI coding agents and blocks potentially destructive actions such as deleting databases or exposing secrets. It integrates with Claude Code, Cursor, Codex and similar agents, giving developers visibility and safety when using AI assistants.
producthunt.com - 13AIDevHunt launch
ForAgent adds private phone numbers for AI agents to call each other
ForAgent provides private phone numbers that let AI agents communicate directly, either on local machines or with external contacts. Calls can be handled via the app or a plugin, supporting agents like Claude Code, Codex, Grok, Cursor, and others. The free tier offers 30 external calls per month, with internal calls always free.
14 points - 14LanguagesDevHunt launch
Rustable launches as a fast, lightweight native database client written in Rust
Rustable is an open-source database client built in Rust, positioned as a lightweight alternative to TablePlus. Its native implementation aims to provide speed and functionality for everyday database tasks, which may interest developers seeking performance-focused tools.
10 points
