One launch.toml.
Six platforms. AI copy scored before you see it.
Pre-flight checks. Timed sequence.
Then it listens for the spike.
Pre-flight runs before a single post leaves your machine. Three severity levels: BLOCK (cannot proceed), WARN (confirm to proceed), PASS. Exit code reflects the worst outcome.
Checks that every configured platform actually has working credentials with write access — before you get 3 hours into a launch sequence and discover your Reddit password changed.
Checks account age and karma against each target subreddit's minimums. BLOCKs if you'd be filtered. WARNs if you're borderline. Saves you from a silent rejection at 6 AM.
HN title scorer (7 signals, 100-pt scale: hype word penalty, personal signal, specificity, tryability). X hook scorer (9 signals). WARN < 60. BLOCK < 40.
Every generated variant is scrubbed (em-dashes rewritten, invisible unicode watermarks stripped) and scored against a 38-pattern AI banlist (delve, leverage, robust, tapestry, "not just X but Y") plus passive voice, vague-word, conversational, and specificity signals. Variants below 50/100 won't ship. WARN < 70. BLOCK < 50.
Manual is a 14-hour day where you forget the timing, post identical copy to two subreddits, and miss the low-competition HN window by 40 minutes.
| hype | Manual | Buffer/Hootsuite | |
|---|---|---|---|
| Pre-launch topic research (HN + Reddit + X) | Yes | No | No |
| AI copy, scored before posting | Yes | No | No |
| Pre-flight credential + karma checks | Yes | No | No |
| Optimal timing window detection | Yes | No | Scheduling only |
| Per-subreddit copy variants (anti-spam) | Yes | No | No |
| HN title scorer (hype-word penalty) | Yes | No | No |
| Resumable on interruption | Yes | No | Partial |
| Auto-post execution engine | Yes | No | Post only |
| Press release + outreach generation | Yes | No | No |
| Post-launch spike monitoring | Yes | No | No |
| Developer tool (OSS-specific) | Yes | No | No |
| You need to be there for 14 hours | No | Yes | Yes |
Not every announcement is a launch. Not every launch needs 14 hours. hype knows the difference.
Before launch day. Pulls HN threads, Reddit posts, X mentions — Claude synthesizes a markdown report: sentiment, top concerns, opportunities.
Full 24-hour blitz. PH, HN, X thread, Bluesky, Reddit waves, Dev.to. Coordinated and timed. Resumable if interrupted.
v2.0, significant release. AI copy, scored review, preflight. X + Bluesky + Reddit + Dev.to.
"Shipped X" tweet. X + Bluesky only. Auto-picks best variant, no interactive review.
Generates press-release.md and outreach-template.md from your project brief. Ready to send to journalists.
Post-launch. Polls HN, Reddit, X concurrently. Alerts on point spikes, milestones, notable mentions.
One config file. The rest is hype.
Interactive wizard sets up platform credentials in ~/.config/hype/config.toml. Never committed to git.
pip install hype-launch && hype config init
Before you write a word. Pull what developers are actually saying about your topic across HN, Reddit, and X. Claude synthesizes a markdown report: top concerns, sentiment, opportunities to address in your copy.
hype signal research "rust CLI tools"
A launch.toml in your project repo. Problem, solution, differentiator, install command. That's what Claude turns into platform copy.
hype config edit launch.toml
hype generates 3 scored variants per platform, runs pre-flight, and shows the scheduled sequence. You confirm once. If you get interrupted, hype launch resume picks up exactly where it stopped.
hype launch new --dry-run
hype launch resume executes the timed sequence — auto-posting to X, Bluesky, Reddit, Dev.to and prompting you for HN and Product Hunt. Use --one to go step by step. Then switch to monitor mode for point velocity, comment surges, and milestone alerts.
hype launch resume && hype signal listen "myproject"
Guided submission where it doesn't. Show HN doesn't have an API, and any tool that pretends otherwise is one Playwright revision from a ban.
| Platform | Read | Write | Mode |
|---|---|---|---|
| X / Twitter | x-reader | OAuth 1.0a | full auto |
| PRAW | PRAW | full auto | |
| Hacker News | Algolia | clipboard | guided |
| Bluesky | AT Proto | AT Proto | full auto |
| Dev.to | REST | REST | full auto |
| Product Hunt | GraphQL | browser | guided |
| Lobste.rs | — | — | manual + copy |
hype signal research <topic> pulls what developers are saying about your space across HN, Reddit, and X — then Claude synthesizes it into a report you can actually use to write better copy.
hype signal listen polls HN, Reddit, and X concurrently. Alerts on point velocity, comment surges, milestone crossings, notable mentions.
Copy this prompt, paste it into Claude, GPT, or any agent. Give it your repo URL. Let it tell you which platforms to target, what the brief should say, and exactly what hype will do on launch day.
Replace [YOUR REPO URL] with your project. The agent reads both READMEs, maps your project to the right platforms, and hands you a ready-to-use launch.toml.
Requires Python 3.12+ and a Claude API key for copy generation.
pip
uv (recommended)
then, in your project
One config, one command, six platforms. hype launched itself with hype. If it couldn't, it would be broken.