Traks is open source. For the past few months I've been building a web analytics platform that runs entirely inside your own Cloudflare account. Today the whole thing is public on GitHub under MIT. Website: traks.dev Code: github.com/shivamanupadi/tra… What it is Self-hosted, cookie-free analytics. You paste one script tag, and every event flows through a Worker in your account into a per-site Durable Object for realtime and into Apache Iceberg on R2 for history. No servers to manage. No third-party service ever touches your traffic data. Not even traks.dev, which only runs the install wizard and forgets you afterwards. Privacy by construction Visitors are counted with a daily-rotating hash, the same approach Plausible uses. Raw IP addresses are never written anywhere. No cookies, no consent banner, no cross-day identifiers. The dashboard never calls out to a third party: auth runs on D1, the world map is self-hosted, the tracker is served inline from the edge. Realtime without a queue Today and live views come straight from a SQLite Durable Object with zero ingest delay. A WebSocket pushes live visitors, pages, referrers, and city-level map dots to the dashboard as they happen. Anything older is answered from Iceberg via R2 SQL, edge-cached and scan-minimized. Built for agents, not just dashboards This is the part I'm most excited about. Your analytics are exposed to AI agents over MCP. Point Claude, Cursor, or any MCP client at your instance and ask "what changed in traffic this week" or "which referrer drove signups yesterday." Bot and agent traffic is classified and reported alongside human traffic, so you can see how much of your audience is now machines. What it costs Nothing, ever, from me. You pay only your own Cloudflare usage, which stays inside the free tier for most sites. A side project runs for about $5/month. 5M pageviews a month is about $7. Past that, roughly $4.60 per additional million events. The full cost model is in the README. Getting started Install in about two minutes at traks.dev. Read the code, file issues, send PRs at github.com/shivamanupadi/tra… If you've wanted analytics you actually own, without running a database or trusting a vendor with your visitors, this is for you. Stars and feedback both very welcome.

You don't have to! That's the point of Traks: it's free and open source. You only pay Cloudflare for usage, which is $0 for most small sites beyond the $5 Workers plan. And you get the cookie-free, no-banner setup Imants mentioned.

Thank you! If anything trips you up on Monday, reply here or open a GitHub issue and I'll take a look.

Thanks! It's one Iceberg table for all your sites, compacted automatically, so the table count never grows. R2 storage is cheap, and the first 10 GB is free. The real cost driver is event volume, and up to ~10M events a month it's about Cloudflare's $5 plan.

Fair worry. That's why it runs on Cloudflare: no server, no database to back up, nothing to patch. Install and updates are one click from traks.dev. If you don't want even that, a hosted analytics tool is the right call.

Agreed. Bots never count as visitors in Traks. They get their own tab, labelled search, AI, SEO, social or monitoring, so GPTBot reading your docs shows up as exactly that, not as a visitor.

Speaking for Traks: there's no catch. It runs in your Cloudflare account, so it costs me nothing per user and I never see your data. No selling data, no ads. If it ever makes money, it'll be a hosted version for people who don't want to run it themselves.

Thanks! Your Cloudflare point applies to Traks too: it only runs on Cloudflare. The upside is that your history is plain Iceberg/Parquet in your own R2 bucket, so the data isn't locked in, and most small sites fit Cloudflare's $5 plan.

Neither: Cloudflare's R2 SQL, their serverless engine over Iceberg tables in R2 Data Catalog. A Worker queries it over HTTP and the results are edge-cached. Today and live views skip it and read from a SQLite Durable Object per site.

Not a n00b question! GA has more depth. What you give up: cookie banners in the EU, your data living with Google, and a heavy script. Traks is cookie-free, the data sits in your own Cloudflare account, and your agent can query it directly over MCP.

감사합니다! Thanks for the thread, and for flagging the $5 Workers plan and the missing revenue feature. That's exactly right.

Both trackers are small. Traks is ~2 KB gzipped (I just measured it). The bigger difference is running it: Umami needs a server and a database you look after. Traks runs on Cloudflare, so there's nothing to patch, back up or scale.

Plausible CE is great. The main difference is what you run: CE is Docker with ClickHouse and Postgres on a server you maintain. Traks has no server. It's Workers, Durable Objects and R2 in your Cloudflare account, installed and updated in one click.

Fair points for a paid product. Traks isn't undercutting anyone, though: it was free and MIT before DataBull launched. There's nothing to charge for, since you pay Cloudflare, not me. And if I disappear tomorrow, the code and your data are still yours.

Thanks for the shoutout, Ekene! I think people will still pay, just for what they can't easily build themselves: hosting, support, someone on call. Traks is free because it runs on your own Cloudflare. If I ever charge, it'll be for running it for you, not for features.

Fair, if revenue by source is what you need, Traks isn't there yet, and I said so in the post. Stripe attribution is next on the roadmap. If you'd like a ping when it ships, I'll let you know.

Glad it's working for you! Analytics is data you should own outright. If it runs in your account, you should be able to read every line of code that touches it. Being open also means people can fix what bugs them instead of waiting on me.

Thank you, that means a lot 🙏 Honest answer: it's built on Cloudflare pieces (Durable Objects, Pipelines, R2 SQL), so a Docker version would mean a rewrite. Running it in your own Cloudflare account is the self-hosting story for now.

Just one: a single events table, with automatic compaction. You pay once, when an event comes in, not every month after. Up to ~10M events a month it's basically Cloudflare's $5 base plan. Stored history on R2 adds cents.

Today and live views never touch R2 SQL. They come from a per-site Durable Object in milliseconds. R2 SQL only handles history, and those queries prune by date and column and are edge-cached. Big date ranges get slower, not more expensive per visitor.