i want to get 2nd place in the @marclou hackathon π₯
behind @indexsy, realistically. but to crack another profit-making business is a W in my book.
so i'm launching https://t.co/CG5jGqFRcQ
π’ i already shipped last night
but was not happy with the design. now I am!
here's what i built and why:
local businesses spend thousands on SEO without knowing whether ChatGPT, Gemini or Perplexity actually recommend them.
and agencies charge hundreds per month to track rankings.
i think both can be done much cheaper.
the idea: pay once for the software. run scans at actual API cost. no subscriptions or inflated credits.
here's what you can do with it:
> π AI visibility grids: see where ChatGPT, Gemini and Perplexity recommend your business across different locations on a map. not just one generic prompt.
> πΊοΈ Google Maps tracking: run up to 15Γ15 grids and see exactly where you rank against competitors.
> π prospect finder: search any service + city and discover local businesses with weak visibility. run batch audits and turn the results into leads.
> π competitor analysis: compare your rankings, reviews, services and AI mentions against the businesses beating you.
> π οΈ AI audit + fix kit: get actionable recommendations, missing service pages, schema markup and a prioritized list of what to improve.
> π white-label reports: generate branded PDFs with heatmaps, findings and shareable links. perfect for agencies pitching new clients.
> π before & after: run another scan after making improvements and show clients what actually changed.
> π€ MCP + REST API: connect your AI agents to automate prospecting, audits and reporting.
the agency use case is what excites me most.
imagine finding 50 dentists in a city, identifying the ones competitors consistently outrank, generating personalized reports and reaching out with actual data instead of another generic SEO pitch.
then charging for the improvements and using the same tool to show the results.
and the economics are pretty interesting.
Google Maps grid scans can cost just a few cents in raw API fees.
so why should you pay $100/month for another dashboard?
i'm starting the lifetime deal at $29.
every 5 sales, the price increases by $10.
you bring your own API keys and pay the providers directly.
pay once, own it for life. no monthly software bill, no maintenance, no headaches.
Higgsfield gave me $100,000 to realize my dream:
A hackathon where participants build a real business, not just an app.
So... Introducing The $100K Hackathon:
π΄ββ οΈ https://t.co/nHDJ5Qruhu π΄ββ οΈ
It starts on November 16, and you'll have 30 days to build a startup and make it profitable. The prizes:
π₯ $100,000
π₯ iPhone Duo
π₯ MacBook Neo
This is not a joke. The winner will get $100,000 in cash.
To win, you must:
1. Start a completely new business on Nov 16
2. Build a digital product (SaaS, directory, etc.)
3. Make the most profit (not revenue)
You can already register at $99. When the hackathon starts on Nov 16, you can still join for $149.
I'm running this with my dear friend @jackfriks, and we'll manually review each potential winner to make sure everything holds up.
See you on November 16, builders π΄ββ οΈπ«‘
nice, Sunil!
would genuinely appreciate an independent security review.
the source code is public, and you can inspect the macOS sandbox entitlements and verify the network restrictions yourself.
happy to send you a free full license, no strings attached.
especially interested in what youβd find questionable or improve. security software should earn trust, not just claim it.
would love to hear your findings!
really interesting approach!
fully agree that agent-level deny rules are a seatbelt, not a true security boundary.
AgentSafe uses local encryption, Touch ID and macOS sandboxing, but an agent with full user permissions can still access plaintext secrets elsewhere.
separating agent and governance permissions at the OS level feels like the stronger long-term approach.
curious how you handle this without adding too much friction to everyday developer workflows?
would love to learn more about your architecture!
my Mac app https://t.co/2hqQuqS8Gv identified 68 leaked API keys out of 181 keys in total π«
i built this app because my API key management was getting ridiculous.
i use Bitwarden for passwords.
but API keys?
some in .env files.
some in random notes.
some in a WhatsApp chat with myself π
and honestly, i hade NO idea how many keys i've created, where they all live, or when i last rotated them.
i just knew that it was getting more and more.
meanwhile, i'm giving AI agents more access to my computer every single day.
Claude Code. Codex. Cursor. Computer use. MCP servers.
they read files, run commands, browse websites and increasingly do things on my behalf.
i love it.
but there's a problem nobody talks about enough:
your AI agent might read an .env file to debug something.
that key can end up in a plain-text transcript.
and potentially get sent to the model provider.
deleting the transcript afterward doesn't undo that exposure.
so i built something for myself.
introducing Agent Safe π
a local-first macOS app for managing API keys and finding the ones your AI agents have exposed.
what it does:
> π ENCRYPTED KEY VAULT
store your API keys locally, protected by Touch ID and a master password.
no cloud account. no syncing secrets to someone else's server.
> π΅οΈ CONTINUOUS LEAK DETECTION
scans Claude Code, Codex, Cursor, shell history, agent configs and .env files.
identifies exposed credentials and ranks them by potential damage.
> π¨ REAL-TIME WARNINGS
a little shield in your Mac menu bar changes color when suspicious keys are detected.
> π§Ή ONE-CLICK CLEANUP
find exactly which files contain exposed keys and remove local copies.
important: if a key already reached a model provider, rotate it. deleting isn't enough.
> π ROTATION TRACKING
track key age and expiry dates, with links to the providers where you can rotate them.
> π€ AGENT-FRIENDLY
use agentsafe run to inject keys into development commands after Touch ID approval, instead of leaving them permanently in .env files.
> π‘οΈ NO NETWORK ACCESS
the macOS sandbox blocks the app's network access.
you can even inspect the entitlement yourself.
and yes, i made the UI look like an old-school Mac.
because i grew up with those interfaces and still love them π
the code is public so you can inspect it and build the app yourself.
the download and basic scan are FREE.
but i added a tiny paid lifetime licence.
π° $0.99 for the first 5 sales.
then +$1 every 5 sales, capped at $5.99.
no subscription. no account.
i'm not trying to build a massive business from this.
i am just super happy using this and wanted to share it, because i think it will solve a real problem for some of you
i just know that there are people who worked exactly like i did, you are exposed and need a fix
and there's something satisfying about someone paying even $0.99 for something you built.
that's real validation.
i'm convinced AI-agent security is going to become an increasingly important problem as we give these agents more autonomy.
this is my tiny attempt to make my own setup safer.
if you use Claude Code, Cursor or Codex on a Mac, i'd love your feedback.
π https://t.co/2hqQuqS8Gv
shipped on Sunday. β
yes, with an important caveat!
AgentSafe adds global Claude Code deny rules for .env files and other sensitive paths.
but the better solution is removing keys from .env entirely.
with agentsafe run -- npm run dev, your app gets the keys from the encrypted vault after Touch ID approval, without storing them in your project.
itβs not foolproof though. deny rules canβt stop every script an agent might execute.
the goal is to reduce exposure, not pretend AI agents with full system access are completely safe.
thatβs why the leak scanner is there too
this is cool! π
shipped AgentSafe just a few hours ago and itβs already getting picked up by indie Mac app reviewers.
started as a solution to my own terrible API key management. now curious whether it actually helps other developers too.
and i love the skepticism. security tools SHOULD be questioned, not blindly trusted.
building in public is getting more fun every day β
Agent Safe (@meestierolff) manages API keys locally for tools like Cursor and Codex. Menu bar warnings and one-click cleanup claimed. Desk has not installed; check isolation yourself first. Pay-once $0.99β$5.99. No account.
@georgeorch thanks George! π
honestly didnβt expect that on launch day
no paid promotion, just sharing a problem i had and what i built to solve it.
β
man youβre basically running my SEO QA for free at this point ππ
I appreciate it tho!
another good catch! adding the canonical tag alongside the JSON-LD fix.
funny how i spend so much time building SEO/AEO tools with RankbyAI and still overlook the basics on my own launch
really appreciate you taking the time to check the actual HTML. this is why i love building in public!
didnt expect to get a free technical audit in the replies to my launch post haha π
great catch on the missing JSON-LD. adding SoftwareApplication schema with the macOS requirements, local-first architecture and security features to the roadmap.
especially important for a security product to make those claims clear and verifiable.
appreciate you actually running the scan! curious if Fecto spotted anything else worth fixing? π
yes!
the funny thing is i probably wouldnβt have built AgentSafe if my own setup wasnβt such a disaster.
turns out scratching your own itch is also a pretty good way to discover that other people have the exact same problem.
now letβs see if theyβll actually pay $0.99 to fix it π
honestly, completely organic!
i just shared the launch and my personal API key disaster on X this morning.
no paid promotion, no outreach to reviewers. @BalloonBoysADV somehow picked it up and featured it a few hours later.
of course I really appreciate it π₯Ί
iβve been building in public for months with almost zero engagement, but lately it feels like iβm finally finding my people.
still figuring out distribution, but sharing real problems and what i build to solve them seems to resonate! β
pretty cool to see AgentSafe getting picked up by indie Mac app reviewers on launch day!
curious if @hii_mohit has spotted it yet π
might be a nice fit for @trymacapps, especially with the nostalgic Mac UI and local-first approach.
would love to hear what you think!
thanks for featuring AgentSafe! π
and fully agree: donβt just trust the security claims, verify them.
the source code is public and the macOS sandbox entitlements are inspectable.
would love for you to actually test it! happy to offer a free full license, no strings attached.
and if you find anything that needs fixing, iβll happily address it publicly.
appreciate what youβre doing for indie Mac developers! π€
great point Jun!
one key can leave traces across multiple .env files, agent transcripts and shell histories.
68 findings doesn't necessarily mean 68 unique keys.
grouping duplicates into one finding with all exposure locations + a single rotation action would make cleanup much easier.
adding this to the roadmap!
@agshinrajabov awesome, looking forward to it! π
especially curious whether the same businesses consistently show up or if the rankings are all over the place.
happy to compare notes once youβve got some results. might make for an interesting little study π
what i shipped today on RankbyAI π’
AI search rankings are messy.
the same prompt can return different businesses depending on the language, location or even which time you ask.
so i built tools to make that measurable.
π consistency checks
run the same AI visibility grid twice to measure how stable the recommendations actually are. optionally test a second language too.
π "why it changed"
compare two scans and see what changed in the recommendations and cited sources.
π€ agent-native workflows
added /SKILL.md so AI agents can discover and use RankbyAI's MCP tools.
π¨ prospecting
built an email-ready teaser showing a prospect's visibility score, competitors ahead of them and a link to their report.
π‘οΈ more transparent AI crawler audits
documented how the audit checks crawler access and what those results actually mean.
but here's the bigger idea:
> Claude Code and Cursor can handle a huge part of on-page SEO now.
technical audits. schema. comparison pages. internal linking.
> yet Google and ChatGPT can still recommend your competitors.
because editing your own website doesn't automatically earn third-party trust.
the next step for RankbyAI:
1. find the prompts your buyers actually use
2. identify the pages AI engines repeatedly cite
3. find which of those pages already mention competitors
4. help you prioritize relevant placements and outreach
not random backlinks.
the actual sources influencing AI recommendations.
that's what i'm building toward.
https://t.co/sRl7bK3WhW π
@carlomakes same here ππ
and thatβs just what the scanner found in my current setup. havenβt even checked git history yet.
the scary part is how normal it feels to keep approving AI agent permissions without ever auditing what they leave behind.
curious what your own Mac would turn up
Shipped! π’
AgentSafe 0.4.0 now checks key files (.env, agent configs, shell history) in every git commit, and whether it was pushed.
My Mac: 69 leaked keys. 9 also in a .env, 1 in git history (pushed), 0 only in history.
Rotating alone kills the key; the history rewrite is cleanup.
right?! π
and thatβs before even checking git history.
the scary part is how quickly you get used to giving AI agents access to everything while barely questioning what they leave behind.
iβm starting to think regular secret scans should be as normal as updating your dependencies.
have you ever scanned your own agent transcripts?
mostly .env files, agent transcripts and configs!
git history isnβt covered by the scanner yet, so the 68 findings donβt include anything buried in old commits π
actually got similar feedback from another dev today. adding a git history deep scan is shipping today!
a little scared what the total will look like after that
exactly!
the tricky part is that the more we trust our AI agents, the more access we casually give them.
meanwhile, mostly our security habits barely change
wanted to make the safer option as frictionless as possible.
curious, have you ever checked what your own agents leave behind in their transcripts?
nice!
honestly, starting manually is a smart way to figure out which prompts actually matter before automating everything.
iβm now experimenting with how many repeat runs are enough to get a reliable baseline without burning through API credits.
would love to compare results once youβve tried it! especially curious how much variation you see between runs π
and the scary part is that i probably wouldβve kept going for months without noticing.
the more autonomy i gave my AI agents, the less attention i paid to what they were leaving behind.
68 findings later, i finally had my wake-up call
curious how many other devs would be equally surprised by their first scan.