Feed aggregator

airpods connect to mac → mic hijacked automatically → you sound like a tin can. mac has a 3-mic built-in array made for calls. airpods doesn't come close.

this runs in the background, keeps airpods as output, keeps mac as input. native swift. ~10mb ram. zero cpu.

``` npx fix-my-mic ```

that's it.

Comments URL: https://news.ycombinator.com/item?id=47097799

Points: 2

# Comments: 0

Article URL: https://github.com/soundspear/formula

Comments URL: https://news.ycombinator.com/item?id=47097794

Points: 1

# Comments: 0

Article URL: https://www.youtube.com/watch?v=da9GwXX-0Zs

Comments URL: https://news.ycombinator.com/item?id=47097792

Points: 1

# Comments: 0

Article URL: https://openclaw.ai/blog/virustotal-partnership

Comments URL: https://news.ycombinator.com/item?id=47097769

Points: 1

# Comments: 0

Article URL: https://www.wired.com/story/inside-rolling-layoffs-jack-dorsey-block/

Comments URL: https://news.ycombinator.com/item?id=47097730

Points: 3

# Comments: 0

Hi HN,

I’m looking for advice on distribution and positioning for a privacy-first AI chat system we’ve been building.

We have two open-source pieces:

- Conduit — a native mobile client for Open WebUI (GPL). It started as a personal project, but now has teams using it and asking for things like SSO, reverse-proxy auth, and security reviews. - Onera — a privacy-first AI chat backend + client where conversations are end-to-end encrypted, with models running inside TEE enclaves. The operator (including us) can’t see conversation content. The client is fully open source and auditable.

The users we’re seeing are: - companies already using Open WebUI who want a mobile client - teams that want a hosted or self-hosted AI chat but don’t want the operator to access message data - privacy or security-sensitive orgs (internal tools, regulated environments, etc.)

What we’re struggling with isn’t building, it’s distribution: - How do teams usually discover tools like this? - Is it better to lead with "Open WebUI ecosystem" or "private AI chat"? - For companies evaluating something like this, what actually matters early on (audits, reference deployments, integrations, etc.)?

If you were building or buying something like this, how would you approach getting it in front of the right users?

Conduit: https://github.com/cogwheel0/conduit Onera: https://github.com/onera-app/onera

Comments URL: https://news.ycombinator.com/item?id=47097729

Points: 1

# Comments: 0

No YAML to write, no docs to read. GitAuto detects your language, framework, and test setup - then opens a PR with the right CI workflow. Merge it, pick which repos to schedule, done.

Already have repos installed? Hit the setup button on the Coverage Dashboard or Charts page.

Comments URL: https://news.ycombinator.com/item?id=47097708

Points: 1

# Comments: 0

Article URL: https://github.com/Unayung/openclaw-browser-relay

Comments URL: https://news.ycombinator.com/item?id=47097692

Points: 1

# Comments: 0

I built VaultLink to solve a recurring problem: how do you securely share an API key or password with someone when you don’t use the same password manager (or don’t use one at all)?

VaultLink encrypts secrets client-side using the Web Crypto API (AES-256-GCM). The encryption key is delivered via a URL fragment (#key=...), which is never sent to the server. The server stores only ciphertext, IV, and salt. Decryption happens entirely in the browser.

Access requires email OTP, and view limits are enforced atomically at the database level to prevent race conditions.

It’s not trying to replace password managers or prevent a recipient from copying a secret. The goal is to reduce accidental exposure and long-lived credential leaks in chat.

Comments URL: https://news.ycombinator.com/item?id=47097666

Points: 2

# Comments: 0

I've accumulated a ton of useful skills (skill.md) for my AI agents and to use on .cursor, .claude., .codex.

But I needed a tool to ensure they would respect some basic rules to avoid security issues and burning too quickly my tokens.

Now, with npx skill-check, I can quickly inspect any online skills or my own and fix any issue automatically.

Comments URL: https://news.ycombinator.com/item?id=47097663

Points: 1

# Comments: 0

Article URL: https://webhookstream.com

Comments URL: https://news.ycombinator.com/item?id=47097627

Points: 1

# Comments: 1

Article URL: https://www.nature.com/articles/s41586-026-10098-2

Comments URL: https://news.ycombinator.com/item?id=47097617

Points: 2

# Comments: 0

Zimbabwe has 90+ towns and cities, a population of 15+ million, significant agricultural and mining sectors — and almost no weather infrastructure built specifically for it. Global apps cover Harare and Bulawayo at best, and the AI summaries they generate read like they were written for someone in London. I built mukoko weather to fix this. A few things that shaped the approach: Weather as a public good. The platform is free, no ads, no paywalls. If a smallholder farmer in Chinhoyi needs frost risk data to protect their crops tonight, that can’t be behind a subscription. Hyperlocal context matters more than raw data. Zimbabwe has distinct agricultural seasons — Zhizha (rainy), Chirimo (spring), Masika (early rains), Munakamwe (winter). Elevation varies dramatically: the Highveld sits above 1,200m, the Zambezi Valley below 500m. The AI assistant, Shamwari Weather, is prompted with this geographic and seasonal context so its advice is actually meaningful to the user. Constrained environments are the primary target, not an edge case. Mobile-first, bandwidth-efficient, PWA-installable. The majority of users are on Android, often on 3G or spotty LTE. That’s not a future concern — it’s the baseline. Technical decisions: ∙ Next.js 15 App Router on Cloudflare Pages + Workers ∙ AI summaries via Anthropic Claude SDK, server-side only, cached at the edge with KV and immutable TTL tiers (AI-generated weather advice shouldn’t change retroactively) ∙ Open-Meteo for forecast data (free, high-quality global model coverage) ∙ 90+ Zimbabwe locations validated against geographic bounds with elevation data for frost modelling The broader vision is weather as infrastructure within a larger Africa super app (Mukoko Africa), with Zimbabwe as the proof of concept before expanding to other developing country markets using the same locally-grounded approach. Would love feedback on the approach, especially from anyone who’s built for similar markets — low-bandwidth, mobile-dominant, regions underserved by global platforms.

Comments URL: https://news.ycombinator.com/item?id=47097611

Points: 1

# Comments: 0

Article URL: https://astiango.com/developers/docs

Comments URL: https://news.ycombinator.com/item?id=47097607

Points: 1

# Comments: 1

WP2TXT is a command-line tool that extracts plain text from Wikipedia dump files. I originally built it in 2006 for corpus linguistics research and have maintained it since. The latest version (2.1) was largely rewritten with features for selective extraction:

- Auto-download dumps by language code (350+ languages) - Extract specific articles by title without downloading the full dump - Extract articles from a Wikipedia category with subcategory recursion - Extract specific sections by name with alias matching (e.g., "Plot" also matches "Synopsis") - Template expansion (dates, coordinates, unit conversions → readable text) - Content type markers ([MATH], [TABLE], etc.) instead of silent removal - Category metadata preserved in output - JSON/JSONL output - Parallel processing (English Wikipedia 24 GB dump: ~2 hours on Apple M4) - Written in Ruby.

Comments URL: https://news.ycombinator.com/item?id=47097593

Points: 1

# Comments: 0

I've been working on filepack, a command-line tool for file verification on and off for a while, and it's finally in a state where it's ready for feedback, review, and initial testing.

GitHub repo here: https://github.com/casey/filepack/

It uses a JSON manifest named `filepack.json` containing BLAKE3 file hashes and file lengths.

To create a manifest in the current directory:

filepack create To verify a manifest in the current directory:

filepack verify Manifests can be signed:

# generate keypair filepack keygen # print public key filepack key # create and sign manifest filepack create --sign And checked to have a signature from a particular public key:

filepack verify --key Signatures are made over the root of a merkle tree built from the contents of the manifest.

The root hash of this merkle tree is called a "package fingerprint", and provides a globally-unique identifier for a package.

The package fingerprint can be printed:

filepack fingerprint And a package can be verified to have a particular fingerprint:

filepack verify --fingerprint Additionally, and I think possibly most interestingly, a format for machine-readable metadata is defined, allowing packages to be self-describing, making collections of packages indexable and browsable with a better user interface than the folder-of-files ux possible otherwise.

Any feedback, issues, feature request, and design critique is most welcome! I tried to include a lot of details in the readme, so definitely check it out.

Comments URL: https://news.ycombinator.com/item?id=47097590

Points: 1

# Comments: 0

Production-ready AI agent that reviews GitHub pull requests using Google's ADK and Gemini 2.5 Flash. Analyzes code for security, performance, and correctness issues. Deploys to Cloud Run, posts structured feedback as PR comments.

Comments URL: https://news.ycombinator.com/item?id=47097588

Points: 1

# Comments: 0

Article URL: https://github.com/sonamsingh25437-ship-it/NF-1-PROGRAMMING-LANGUAGE

Comments URL: https://news.ycombinator.com/item?id=47097570

Points: 1

# Comments: 0

Article URL: https://dixken.de/blog/lets-burn-some-tokens

Comments URL: https://news.ycombinator.com/item?id=47097569

Points: 1

# Comments: 0

Article URL: https://fastfilelink.com/static/blog/ai-fatigue-test-only-zero-code-review.html#ai-fatigue-test-only-zero-code-review

Comments URL: https://news.ycombinator.com/item?id=47097560

Points: 1

# Comments: 0