Yesterday's Top Launches: 5 Tools from September 28, 2026
Four of yesterday's five launches share the goal of capturing activity as it happens instead of reconstructing it manually later, with most built on files and feeds users already own.
Five launches landed yesterday, and four of them circle the same idea from different angles: capture what actually happened while it’s happening, because reconstructing it later by hand is the part nobody does. For anyone tracking new developer tools, the useful signal here isn’t the count, it’s how many of these are built on files and feeds you already own.
Hemory
The name is Hear plus Memory, and that’s the whole product: an always-on listening app that turns your day into a private, searchable memory store your AI agents can query over MCP. The distinction from meeting transcription matters. A release review gets captured either way, but the lunch after it, the voice note on the walk back, and the small commitments made in passing usually evaporate. Hemory listens on the phone or Apple Watch you already carry, splits your day into speaker-labeled moments, and exposes a single search_memory tool to Claude Code, Codex, Cursor, VS Code and similar clients.
Two listening modes keep the privacy boundary yours: manual, and scheduled windows like weekdays 9:00 to 18:00. Quotas are counted by voice activity detection rather than wall-clock time, so silence and background noise don’t burn your hours. Audio is processed as a stream and never stored in the cloud, with a self-host option listed as coming soon. Pricing runs from a free 10-hour trial to $10/20/60-hour tiers and $50 for unlimited.
Worth flagging: the quotas are measured in “VAD hours,” which is a reasonable unit and a confusing one to compare against anything else. And the payoff depends on your agents being good enough to use that memory well. Still, the demo of asking an agent what a colleague promised in Tuesday’s sync, and getting the launch budget commitment back, is the most convincing argument for ambient capture I’ve seen.
Chit
A 777 KB Mac app with no network code at all. Chit reads the Claude Code transcripts already sitting at ~/.claude/projects, groups everything by repository with your busiest project first, and prints the day back as a receipt you paste into standup or a timesheet. Worktrees fold into their parent project. Lines are built from session titles or changed file names, never your prompts, and output shows audit.js rather than a full path.
There’s a CLI with --today, --date=, --from=/--to= and --markdown flags, so you can wire it into hour tracking. Free forever covers today’s receipt; Pro is $12 once for past days, week and month statements, and a dashboard. Scan speed isn’t a problem — 2.7 GB of transcripts takes about a second. The catch: it isn’t notarised yet, so a browser download trips Gatekeeper once. The terminal installer doesn’t.
Paragraph Notes
The quietest launch of the five. A Mac App Store Markdown notes app that collects no data, never connects to the internet, and stores notes locally as plain text. Free up to 20 notes, $19.99 one-time for unlimited.
The pitch is honesty about where your writing lives, and that’s a real concern for a lot of people. But the feature set is thin — a distraction-free plain text editor is not hard to find, and free options exist. What you’re buying is the App Store install, the absence of network code, and the portable format. If you don’t care about those three things specifically, this isn’t for you.
GoodSocials
An AI LinkedIn posting system with a human approval board. Sign in with LinkedIn, paste your website, and seven posts appear in 90 seconds. Posts come in three kinds: market research, deep dives, and your own numbers pulled read-only from PostHog, Stripe, GitHub, Plausible and Notion. Corrections become persistent voice rules, so a note like “open on the number, not a question” applies to every post after it.
Pro is $100/month for one profile, Agency $1,000 for ten, Consultancy $2,000. There’s a genuine thesis here about avoiding the cringe openings that make AI-written posts recognizable. The jam-study post comparing Iyengar & Lepper against a 50-experiment meta-analysis is a good example of the standard they’re aiming for — though I’d note that “grounded in data” is itself a genre, and a board full of it can read as formulaic in a different way. Treat the $3,000/month manager comparison as marketing arithmetic rather than a benchmark.
Psst
Shared lists for two people or twenty, joined by a single link. Add items by typing naturally (“milk tomorrow 6pm” becomes an item with a date, time and reminder), assign them, and watch everything sync live. Offline works, merging edits when you reconnect.
The genuinely interesting part is price memory. Photograph a receipt and the lines come back structured, each with a crop of the row it came from and what that item cost last time. Comparisons use unit price, not line totals, so buying 2kg instead of 1kg isn’t reported as a price rise, and it stays quiet when two prices can’t be compared safely. That restraint is the right call, and rarer than it should be. Free tier covers shared lists; Pro is $35.99/year, Family $49.99 for six, Team $79 per user.
Community voting is still open on yesterday’s batch, so no rankings have settled yet. If forced to pick, Chit and Psst have the tightest scope-to-value ratio, and Hemory has the highest ceiling.
Quick links
- Hemory — always-on listening memory for AI agents
- Chit — daily receipt from your Claude Code transcripts
- Paragraph Notes — local-only Markdown notes for Mac
- GoodSocials — researched LinkedIn posts with an approval board
- Psst — shared lists with honest receipt-based price history