Yesterday's Top Launches: 5 Tools from October 18, 2026
A roundup of five unrelated launches in launch order, from a browser chess drill and cold-gas-thruster electric car to an e-signature tool, a Base44 chat app, and a Mac app for managing coding agents.
Yesterday’s slate was unusually scattered. A browser chess drill, an electric car with cold-gas thrusters, an e-signature tool that refuses to sell subscriptions, a chat app built on Base44, and a Mac app that herds coding agents. Five launches, five completely different audiences. For anyone watching new developer tools specifically, the last two entries do most of the work here, though the rest are worth a look for other reasons.
A note on ordering: community vote counts weren’t available for any of these at publish time, so nothing below is ranked by popularity. The picks are described in launch order.
Chathonk
A chat platform for friends, with web games and audio calls bundled in. You sign in with Google or an email and password, or skip all of that and continue as a guest, which is a nice touch for anyone who just wants to poke around before committing. Once inside, you find people by email address or username, then talk. Emojis and GIFs handle the expressive side.
That’s roughly the whole product description, and it’s thin. No word on which games are included, whether guest users can join a call, or how group conversations work. The footer’s “Edit with Base44” label suggests it was assembled on a no-code platform, which explains the sparse feature list and probably the free price. If your friend group already lives in three different messaging apps, this won’t replace any of them. If you want a small, casual space with a game or two built in, it’s worth ten minutes. Just don’t expect Discord.
Tesla Roadster
Tesla’s page leans hard into spectacle: three trims called Performance, Thrust and Rocket, active aero with a suction fan system that supposedly holds the car to the ground while stationary, and cold-gas thrusters that cut 0-60 to under a second in the Thrust build. The Rocket trim flies. It also isn’t approved for public roads, is invite-only, and isn’t sold in every region.
The engineering claims are genuinely interesting — steer-by-wire, magnesium and carbon fibre construction, a fixed seat position to shave mass. The practical details are not. No pricing, no delivery timeline, and a footnote admitting the specs shown reflect the Thrust augment as an optional upgrade. Reserve links exist; what happens after you click one is unclear. Treat this as a concept car with a configurator rather than something you’ll be driving.
SignWith
E-signatures priced per document instead of per month, aimed at people who send a handful of contracts rather than hundreds. The pitch is blunt: a $20/month plan costs $240 a year, while someone sending five documents monthly pays about $19 with the free tier plus an occasional credit pack.
The workflow is the standard four steps — upload, add signers, place fields, send — and the extras are all included on free accounts, which is the part worth noticing. Email OTP verification, audit trails with timestamps and IP addresses, certificates of completion with SHA-256 fingerprints, signing order, reminders. Credit packs run $9 for 10, $19 for 25, $29 for 50, with a $199 lifetime option. There’s a REST API and an MCP server with 18 tools for AI assistants, both drawing on the same credits.
Where it gets shaky is at scale. Fifty documents a month means buying packs repeatedly, and at some volume a subscription genuinely wins. But for a consultant sending two statements of work a week, the math is hard to argue with.
Memory Chess
A free browser game that tests whether you can hold a chess position in your head. You study a board for a set time, it blanks, and you rebuild it. You get accuracy from 0 to 100% plus a square-by-square map of what you missed. Piece count runs 2 to 32, study time 1 to 60 seconds, and the site claims nearly 49,000 games played.
The clever part is the framing. Most chess training drills decisions; almost none checks whether your mental picture of the board is accurate, which is exactly where calculation falls apart three moves deep. There are calibration rounds, four difficulty presets, a custom rig, a daily board shared by everyone, and spaced review of positions you scored under 80% on. The Learning Center ties each drill to a guide on board vision or blindfold play.
Honest caveat: there’s no account and no server. Your stats live in that browser, and cross-device backup is listed as coming soon. Switch laptops and you start over unless you remember to export.
Parallax
The most substantial launch of the day, and the one with the clearest problem statement. Running several coding agents in parallel multiplies output, but it also multiplies the chances one drifts from what you asked, and reviewing the results is its own bottleneck. Parallax puts Claude Code, Codex, Cursor, Gemini CLI, OpenCode and others in one Mac app, with a layer called the Captain that splits a single request into tasks, checks each step against your original ask, and returns pull requests.
The review claim is the interesting one. On Martian’s Code Review Benchmark, Parallax found 2.24 merge-blocking comments per pull request against 1.38 for the next tool listed. It’s a 50-PR dataset, so treat the number as directional rather than definitive, but the mechanism behind it makes sense — reviews run against the full session, not just the diff. Subscription fallback when one Claude plan hits its weekly cap is the other feature that will matter to anyone paying for two of them.
Free to start with $50 of credit and five reviews, GitHub for sign-in and repos, a notarised .dmg for Mac. Linux users get a headless daemon on port 7878, no native window.