Hi, I'm Sai Akash Neela 👋

Every indie product I run had the same blind spot. My analytics told me how many people visited, and Stripe told me how much money came in, but nothing connected the two. I could see a spike from Hacker News and a spike in sales in the same week and still not know whether one caused the other.

I didn't want to bolt another cookie banner onto every site either. Most of my projects are small, privacy-first tools, and asking visitors to accept tracking cookies just to read a landing page felt wrong.

So PilotFox does both jobs in one place. It is cookieless web and product analytics: add one script tag, connect Stripe, Dodo Payments or Polar with a read-only key, and every payment is linked back to the visit, channel and campaign that brought the buyer in. On top of that sit funnels, goals, retention, session replay, web vitals, Google Search Console data and affiliate tracking with ?ref= links.

Because I live inside AI tools all day, PilotFox also ships a REST API with an OpenAPI document and an MCP server, so I can ask an assistant which channel brought paying customers this month instead of clicking through dashboards. It even breaks out traffic from AI assistants such as ChatGPT and Perplexity, which is now a channel worth measuring on its own.

The goal is simple: find the traffic that pays, and stop guessing.

My MacBook kept running out of space, and it was never my photos. It was Xcode DerivedData, browser caches, the npm cache and app logs quietly piling up from years of builds.

The commercial cleaners I tried didn't sit right with me. Many are heavy Electron apps with background daemons and subscriptions, and I don't love giving an opaque tool permission to delete things on the machine I work on.

So I wrote MacKitty in 100% native Swift and SwiftUI, as a universal binary for Apple silicon and Intel Macs. It only ever cleans a fixed list of rebuildable cache and log folders. Documents, Desktop, Downloads, Photos, Git repositories, the Keychain and macOS system files are never touched, scans are a dry run, and nothing is removed until you review it.

It also gives you a one-screen dashboard with storage, live CPU and memory, battery and network, plus a menu bar popover. The built-in updater only installs builds signed with MacKitty's Developer ID and notarised by Apple. No telemetry, no ads, no subscription.

It's free and the source is on GitHub. Allow Kitty to take care of your Mac.

I spend a lot of time on Cloudflare Workers, and I kept finding brilliant open-source apps built for it scattered across GitHub: URL shorteners, uptime monitors, temporary email, analytics, even full CMSs. There was no single place to discover them.

Meanwhile I was paying monthly for SaaS that could run in my own Cloudflare account, often on the free tier, with my data staying with me.

OpenFlareStack is the catalog I wanted. It lists open-source, self-hosted apps that deploy to your own Cloudflare account, discovered from GitHub and re-ranked every night by stars and recent commits. Every listing is open source, not a fork, not archived, and has had a commit in the last 12 months. Most have a Deploy to Cloudflare button that copies the repository into your GitHub account and creates the D1 databases, KV namespaces and R2 buckets it needs.

It also maps popular swaps, such as OpenSEO instead of Semrush, Sink instead of Bitly and UptimeFlare instead of UptimeRobot, so you can start from the bill you want to cancel. OpenFlareStack never hosts anyone's code. It links straight to each author's repository, and the ranking logic is open source too.

TopTheBid started as a weekend experiment and a bit of a joke: what if a leaderboard had exactly one rule, and whoever pays the most sits on top?

There are no accounts and no sign-up. You enter your X handle, your website and an amount from $1, pay by card through Stripe, and the listing goes live the moment payment clears. To take a spot you bid at least $1 more than the listing in it. Bids only go up, ties go to whoever got there first, and placements never expire. They only move down when someone pays more.

I built it to play with a tiny, honest product: one page, one payment flow and rules simple enough to fit in a tweet. It was also a fun way to put my own projects on the board.

It's paid placement, not an endorsement, and it says so clearly. Sometimes the best way to learn about pricing, payments and virality is to ship something small and watch what people do with it.

Small businesses miss calls. A clinic, a salon or a local shop can't staff a phone line around the clock, and every missed call is a customer who rings the next place on the list. Those same businesses rarely have the budget for a proper video ad either.

I built ArivuLabs to solve both from one input: your website. Paste the link and ArivuAgents builds a voice agent that has read your pages and answers calls in 70+ languages, with accent-native voices for Indian languages, so Tamil sounds Tamil and Malayalam sounds Malayalam. It qualifies leads, books appointments, handles interruptions naturally and hands off to a person when judgement is needed.

ArivuStudio is the creative side: video generation, image and ad creation, multilingual dubbing, voice cloning and speech, with the cost of every run shown up front.

Building it meant real-time speech, telephony and a lot of care around the rules. Outbound calls are only placed within legal calling hours, always open by saying the caller is talking to an AI, and skip anyone on the business's do-not-call list. Every new account gets 100 minutes and a phone number free for 4 weeks, because the best demo is your own phone ringing.

The Bhagavad Gita is a text I keep coming back to, but most places to read it online are cluttered with ads, split across languages, or hard to read on a phone.

Gita Sadhana is the study companion I wanted: all 18 chapters and 701 verses in Sanskrit with transliteration, a clear translation, a reflection and the essence of every verse, in 12 languages including Hindi, Telugu, Tamil, Kannada, Malayalam, Bengali and Marathi. Every verse has its own page and its own recitation.

I added small habits that support a daily practice: a verse of the day, saved verses and personal reflections, a reading streak, and Gita AI for asking questions about the teachings.

It's free. As an engineer, it was a lesson in building for readers rather than users: fast pages, good typography and structured content that search engines and AI assistants can understand verse by verse.

I honestly dislike the AWS console for day-to-day file management. It's slow and multi-layered, and finding one nested object feels like searching for a needle in a haystack. I also split my week between macOS, Windows and Linux, so a Windows-only S3 browser was never going to cut it.

BucketStack is a free, open-source (MIT) desktop app that makes S3 buckets feel like local folders. It's built with Rust and Tauri 2 instead of Electron, with a React 19 and TypeScript interface, so it stays lightweight.

It's not locked to AWS. I use it daily with Cloudflare R2, MinIO, Wasabi, Backblaze B2, DigitalOcean Spaces and Railway. You can copy files directly between providers with live progress, edit JSON, YAML, Markdown and code in the built-in editor, restore deleted files from Trash, share expiring links, back up folders on a schedule and see storage breakdowns.

Access keys stay on your computer, encrypted with AES-256-GCM. No account, no tracking. Builds are available for macOS (Apple silicon and Intel), Windows 10 and 11, and Linux (AppImage, .deb and .rpm).

Half the thoughts in my working day are throwaway: a command I need in five minutes, a name for a variable, a to-do I'll forget by lunch. Opening a full note-taking app for that felt like overkill, and most of them want an account before you can type a word.

StickyDraft is a sticky-notes board that opens in your browser and is ready the moment it loads. Click anywhere to start a note, drag and resize notes, pick from six colours, split work across multiple boards and switch to dark mode.

Notes are saved locally in your browser, so it works offline, with no signup and no third-party analytics, ads or tracking. If you want your drafts on more than one device, you can optionally sign in with Google or GitHub to sync them.

The source is on GitHub. Sometimes the best tool is the one that gets out of your way.

Every sign-up form I build eventually needs the same thing: catch typos, block throwaway inboxes and stop fake accounts without punishing real users. The usual answer is a paid validation API that bills per check and sees every address you send it.

I didn't want either of those, so I built checkemail, a free, open-source (MIT) email validation engine with zero runtime dependencies.

It runs four checks, cheapest first. Syntax, 9,000+ disposable domains and role inboxes like support@ are checked locally with no network call. Then DNS confirms the domain exists and has MX records. Then it fingerprints the provider (Gmail, Outlook, Proton and more) and canonicalises aliases. Finally every signal adds up to a 0 to 100 confidence score with a clear status such as LIKELY or DISPOSABLE.

You can npm i @isan3/checkemail in Node.js or Bun, run it as a Docker HTTP API, or use it right in the browser. Only the domain is ever looked up, so addresses never leave your servers, and the disposable list is refreshed every Monday and Thursday by a GitHub Action. It deliberately never sends SMTP probes: it proves a domain can receive mail, not that a specific mailbox exists.

Signing a rental agreement, merging bank statements, redacting a passport scan: these are exactly the documents I don't want to upload to a random website just to rotate a page.

Browsers can do this work now. PDFamaze is a free, open-source set of 51 PDF tools that run entirely inside your browser tab using pdf-lib, pdf.js and WebAssembly. Files are processed on your device and never uploaded, with no account, no watermark and no usage limits.

It covers the everyday jobs, like signing, filling forms, merging, splitting, compressing, rotating and adding page numbers, and the careful ones: OCR for scanned pages, redaction, removing scripts and embedded files, encryption, changing permissions and even showing any JavaScript hidden inside a PDF.

There's no history by design. Close the tab and it's gone.

The Model Context Protocol (MCP) fundamentally changes AI assistance. I built a specialised MCP server that bridges Claude Desktop directly with Cloudflare Workers and Prometheus metrics.

Instead of opening five browser tabs to check error spikes or toggle routing rules, I now ask: “Inspect error rates on the auth worker over the last 30 minutes.” The agent pulls real-time telemetry, spots the anomaly and proposes the exact fix.

For years I insisted on deploying everything onto production-grade Kubernetes with Helm charts and ingress controllers. That's essential for large enterprises, but for independent builders and small services it adds an enormous cognitive tax.

I moved my personal stack and side projects to Coolify on dedicated Hetzner nodes. The developer experience is stellar: automatic wildcard SSL, instant GitHub PR preview environments and raw Docker Compose simplicity, without losing control of the infrastructure.