Coding agents get the attention, but a steady share of Show HN launches are aimed at ordinary life: the phone call you did not want, the bank statement you have not opened, the appointment where you forgot every question. These projects tend to be small, specific and unusually honest about scope, which makes them easier to judge than the grand platforms.
Here are twelve from the last few weeks, grouped by what they take off your hands.
The phone
Robocalls are the most universally loathed surface in consumer technology, and CallSitter is the most direct response to them: an AI that answers your phone so scammers get their time wasted, sales calls get a polite no, and real callers get their message taken. Calls divert to a telephony number where a live voice agent handles them. What sets it apart is evidence rather than claims: recordings of real scam, spam and genuine calls are published, including a thirteen minute scam call, so you can hear exactly how it behaves before you forward anything.
On the hardware side of voice, Babytalk is a fully offline speech-to-text and text-to-speech system for ESP32-S3 and P4 boards under Micropython and AtomVM, with a model finetuned for noisy environments and 4- and 8-bit integer kernels for Xtensa and RISC-V vector units. It also ships the data generation, field capture and finetuning framework, which is the part most embedded speech projects leave out and the part you eventually need.
Music is the friendlier end of audio. FlexChords pulls guitar chords, key and tempo from any YouTube video or audio track so you can play along with a synced chart, and reports 92 percent chord accuracy on unseen songs measured against the MIREX standard. Publishing the measurement standard is what separates a real claim from marketing.
Money and privacy
Financial data is the thing people are least willing to upload, and Maali is built around that refusal: a personal finance app whose AI analyst runs on your own computer, working with the Wi-Fi off because the data never leaves the machine. You drop in bank statement files or connect once through SimpleFIN, and it sorts every charge and reports on the month.
Privacy shows up again in Smart Blur, a Chrome extension that blurs emails, phone numbers and card numbers on any page as it loads, before you share your screen. It runs on the device with no account, and the paid tier turns blurring on automatically when a screen share starts. Screen sharing is where most accidental data exposure happens, so automating it at exactly that moment is the useful part.
For finance work with audit requirements, Ekselio takes a different position: an AI designs the workflow while a deterministic engine executes it, producing auditable SQL and Excel models with Power Query that recompute and verify their own numbers. Your files are never uploaded because every step runs in the browser. Generated analysis you cannot check is not analysis, and this one is built to be checked.
Health and attention
OpenHand Sidekick joins a telehealth visit in real time, explains medical terms as they come up and suggests questions worth asking. Timing is the whole product: after the appointment is too late to ask anything.
Titan is a desktop assistant that sees what is on your screen and works in the apps you already use, remembering what you tell it between sessions. Screen-level assistance is the obvious next fight after chat-level assistance, and the obvious privacy trade-off comes with it.
Reading, research and place
Documents are where AI quietly earns its keep. Sezwhere answers the question where does it say that: paste text or upload a PDF, Word file or Markdown, ask in plain language, and it highlights the passages that answer you even when the document uses different words. Returning the passage rather than a summary is the right default when someone has to rely on it.
For academic work, Yila searches real papers across OpenAlex, PubMed, arXiv and Wanfang and covers scoping, search, triage and drafting, producing cited reviews, annotated PDFs, per-paper evidence cards and posters. OpenLume goes the other way and answers a technical question with a narrated explainer video you can pause to ask follow-up questions, which suits people who learn by watching rather than reading.
Travelling gets its own entry in Mira, an AI audio tour guide that narrates what is around you as you walk, with a street view demo of San Francisco and Paris that needs no sign-up before you install anything.
What these have in common
Almost every project here makes a deliberate constraint visible: the data does not leave the device, the answer comes with the passage it came from, the accuracy figure comes with the standard it was measured against, the demo works before you sign up. That is a more useful signal than any feature list, and it is why this batch is worth a look even if none of it is aimed at your job.
