Open Source Radar
New open source projects, releases, libraries, developer tools, and important events around free software. What is worth trying and why it matters.
Latest publications
TurboFieldfare makes local AI a storage problem worth testing
An Apache-2.0 Swift/Metal runtime is testing a sharper idea for local LLMs: stream the routed experts a MoE model needs instead of keeping the whole model resident in RAM.
Codex Security is open source, but maintainers should read the small print
OpenAI’s new security CLI looks useful for OSS review, but the first public day exposed access, cost, guardrail and package-name lessons.
PGSimCity turns PostgreSQL internals into a city worth exploring, carefully
The new Apache-2.0 WebGL project is a rare kind of database teaching tool: immediately tryable, visually strong and honest about its prototype limits.
Codeberg draws a line against vibe-coded open source
The independent FLOSS forge says it will not train LLMs on user data and is moving to reject projects mostly built by heavy LLM use. The hard part is what happens next.
curl’s month off from vulnerability reports is a warning for open source
curl paused vulnerability intake for July 2026 after months of pressure from report overload. The real issue is the new economics of AI-shaped security submissions.
LocalSend is the open-source AirDrop alternative we still need
LocalSend is not magic AirDrop for every situation. It is something more useful for mixed-device homes and teams: a private local file-transfer tool that exposes how unsolved nearby sharing still is.
OpenMandriva shows the open source risk hiding behind maintainer access
The alleged repository sabotage is not just distro drama. It is a reminder that open source projects need access control, backups and offboarding before trust breaks.
Bun’s Rust rewrite is not just a Rust versus Zig story
Bun turned a million-line language migration into a public test of AI-assisted open source engineering. The useful lesson is about tests, CI and human ownership.
Rowboat shows where open-source AI work apps are heading
The local-first AI coworker drew Hacker News attention because it turns the next open-source fight into a trust question: who controls work memory?
Organic Maps, CoMaps and the trust problem behind open-source navigation
A fork of Organic Maps has turned an offline map app into a broader test of open-source trust: code licenses, binary map data, infrastructure, finances and governance all matter.
sqlite-utils 4.0 shows how AI can help an open source release without replacing the maintainer
Simon Willison's sqlite-utils 4.0 release candidate is a better AI coding story than hype: public PRs, tests, changelog work, a second model review and a maintainer still owning the decision.
git-annex and the new dependency question: can open source avoid AI-generated code?
Joey Hess spent about 100 hours trying to keep git-annex buildable without LLM-generated dependencies. The result is not a purity tale, but a practical warning about provenance, maintainability, and trust in modern open-source supply chains.
