About

What is this?

AT Proto Collection Tracker is a live dashboard that watches the Jetstream firehose — a real-time stream of every record written to the AT Protocol network — and counts the record types (called collections) it sees. The results are grouped by namespace and updated continuously.

Why does it exist?

The AT Protocol (ATmosphere) is an open, federated social network protocol. Anyone can build an application on it and define their own record schemas. There is no central registry of what applications exist or how active they are.

This tool takes a bottom-up approach to discovery. Rather than starting from a list of known apps, it simply watches what records are actually being written to the network. If a new application launches and its users start creating records, this tracker will surface it — sorted by when records were first seen — even if nobody has announced the app anywhere.

The goal is to make the ATmosphere ecosystem more visible: to reveal the breadth of activity happening across the network, highlight emerging namespaces, and give developers and curious users an honest picture of what is actually being built and used.

How does it work?

A relay service connects to the Jetstream WebSocket and processes every commit event on the firehose. For each record written anywhere on the network it extracts the collection name (the NSID — e.g. social.attic.post) and increments a counter. Counts are flushed to Postgres every few seconds.

Collections are grouped by their two-segment reverse-domain prefix: social.attic.post and social.attic.like both belong to the social.attic namespace, which corresponds to the application at attic.social.

The relay also resolves each user's DID to their Personal Data Server (PDS) in the background at around 50 DIDs per second — without ever blocking the firehose event loop. This lets the tracker show how many different hosting providers have users writing records of each type.

Caveats

• Counts reflect activity observed since the tracker started, not the total ever on the network.
• The First seen timestamp is when this tracker first observed a record of that type, not necessarily when the collection was created.
• High-volume namespaces (e.g. app.bsky.*) can be filtered out using the toggle on the main page.
• PDS attribution lags by a few seconds to hours depending on how recently a DID was first seen.