The developer-facing launches get the upvotes, but they were only half of this week. The other half went at the ordinary things people actually want an assistant for: the group chat that never settles on a plan, the tutorial video where the important part is on screen and not in the words, the anime nobody has recommended to you yet, the job listing that turns out to be fake, and the paragraph you wrote that sounds like a robot.
An assistant that lives in the group chat
Boba is built around a premise most assistants ignore: the hard part of coordinating people is not talking to one person, it is a group that cannot agree. You text it or open it on the web, free, with no app to download and no sign-up, and it works one-on-one or inside your group chats.
The group chat is where it earns its keep. Drop it in the trip chat and it researches cabins, flights and trains, splits the cost, and lays out options everyone can weigh in on, so the decision happens in the thread instead of across six side conversations. Add it to the family chat and it becomes the shared memory: grocery lists, school pickups, appointments, who has Mia on Thursday, with a reminder to the right person at the right time. It also keeps the standing plans going, checking who is in for weekly pickleball or board game night, finding a time and a place, and reminding everyone the day before.
The one-on-one side is more conventional but competently so. Connect Gmail or Outlook and it flags what matters, drafts replies and puts plans on your calendar. Meeting reminders and check-ins on the medication or the appointment you keep pushing follow up so you do not have to. Name the place and the time and it books the table once you say yes, with the confirmation landing in your email.
Two honest caveats from the site itself. Messages you send may be used to improve Boba, and Boba is an assistant that can get things wrong. Texting it also means handing over a phone number, which is the price of the whole design.
Eyes for your AI on YouTube
FrameCap fixes a specific, maddening gap. When you ask ChatGPT or Claude about a YouTube video, it reads the spoken transcript and nothing else. It cannot see the screen. So when the presenter says "just type this exact line into your box" and types a formula without reading out the brackets and commas, the model has no idea what was on screen, and you are rewinding five times to catch five seconds of the thing you needed.
FrameCap is a free Chrome extension that gives your AI eyes. One click on any YouTube video captures two things at once: a timestamped transcript read from the video's captions, and a sharp frame every few seconds, with the interval yours to choose between 1 and 15 seconds. It then hands both to ChatGPT, Claude, Gemini or Perplexity as a PDF with a ready prompt already attached, or downloads them as a ZIP for AI coding agents like Claude Code, Codex and Cursor. Frames go up to 4K, sharp enough to read code on screen.
Everything runs inside your browser, on Chrome, Edge, Brave and Arc, with no account and no API key, which is the part that makes it easy to actually use rather than to plan to use. The limits are structural rather than accidental: it is YouTube only, the frame interval is fixed at capture time, and the transcript comes from the captions, so an uncaptioned video gives you frames and no words.
An anime companion that knows what is airing
Otaku AI is a small, fan-built tool with one idea done well: turn a vague watch-mood into a shortlist you can act on. It is free in the browser with nothing to install, and the site is upfront that it is an independent fan project, not affiliated with, endorsed by or connected to any studio or creator of any anime.
The recommendation side works from mood rather than metadata. Ask it about currently airing shows, or pick sad, hype, chill or dark, and it filters by your own taste. The character side is the part people will actually spend time with: Character RP lets you talk to or roleplay as characters from a series you love as if they were really there, all through plain conversation. Around those sit a set of tools that read like a fan's own checklist: anime search, a character creator, an anime quiz, deep dives, power-scaling arguments and a philosophy chat.
Conversations stay lightweight, nothing you type is shared with anyone else, and you can share a response as text or export a whole chat as a downloadable card image to post. Login is required to save chat history. It is narrow by design and useful only to anime fans, which is a better trade than being broad and useless.
A job search that checks for ghost jobs
Four-Leaf attacks a problem everyone who has job-hunted recently knows: a chunk of the listings you find are not real. It scrapes over 100,000 live openings directly from company career pages with ghost-job checks built in, and the search itself is free with no account needed.
Past search, it is a stack of seven assistants meant to carry you from listing to signed offer. Voice mock interviews, a resume builder, cover letters, a compensation negotiation coach, an application tracker, an email assistant, job discovery and a LinkedIn optimizer, plus a set of free standalone tools including a resume checker, a resume tailor, a cover letter generator and the ghost job checker on its own. Hiring data, live job market trends and interview guides with question banks organised by role sit alongside, and there is an MCP server that puts Four-Leaf inside Claude, Cursor and ChatGPT.
The split between free and paid is worth knowing before you start: the search and the standalone tools are free, and the seven assistants sit behind a paid trial. Coverage is also limited to employers whose career pages it can scrape, so it is a wide net rather than a complete one, and being a web product there is nothing that works offline.
A linter for prose that reads like AI
AiSniff closes the loop on the rest of this list. If you have used an assistant to draft anything, you have also read a paragraph that opens with "in today's fast-paced world" and calls something a game-changer. AiSniff is an open-source linter that underlines phrases in English and Chinese text matching documented AI-writing patterns. Each mark names its rule, cites a public source and suggests a plain rewrite, all from one set of JSON rules with no server and no model call. The web editor, the CLI (`npx github:Arthur031221/AiSniff lint draft.md`) and the agent skill all read the same rules.
What makes it worth trusting is that it publishes its own failure modes. On answers held out from HC3, a public set of questions with human and chatbot answers, the English rules marked at least one phrase in 29% of chatbot answers and 2% of human ones; for Chinese it was 24% and 6%. But 71% of the English chatbot answers had no hit at all, the AUROC of hit density is 0.63 for English and 0.59 for Chinese, and several rules fire as often on human answers as on chatbot ones. The docs list them. It states outright that it reports patterns and says nothing about who wrote the text: a clean result does not prove a person wrote it, and a hit does not prove a model did. MIT licensed, and it is better read as an editing aid than as a detector.
The everyday pattern
Put the five side by side and a pattern shows up that has little to do with capability. Boba, Otaku AI and Four-Leaf all win by being where the task already happens, in the group chat, in the browser, on the careers page, rather than asking you to come to a new app. FrameCap wins by fixing one missing sense instead of adding a feature. And AiSniff wins by being honest about what it cannot do, which is the rarest launch strategy of the five.