How to build your own PDS

This is a long-form, hands-on guide to building a Personal Data Server for the

A Personal Data Server is the piece of the AT Protocol that holds your

Before we touch any code, we need shared vocabulary. The AT Protocol is built

This chapter is the map. It explains how the code under [src/](../src/) is

Identity is the foundation. Before anyone can store a single byte on this

Before we can build the Merkle Search Tree in the next chapter, we need to

This is the chapter that pays for the rest of the book. The Merkle Search

The MST from the [previous chapter](./06-merkle-search-tree.md) gives us a

By chapter 07 every block in the repository is content-addressed: a chunk

Every record on this PDS, every XRPC procedure it serves, every event on

The AT Protocol's RPC layer is called XRPC. The name promises something

The PDS's authoritative storage is Postgres. The MST itself isn't in

This is the longest single endpoint in the PDS. Registering an account

Account creation handed back two JWTs. This chapter is about what those

Chapter 12 created the account and an empty repo. Chapter 13 made the

A typical Bluesky post is a few hundred bytes of JSON. A single image

The firehose is the PDS's public stream. Every commit on every account on

You've built a PDS. By chapter 16 it owns repositories, signs commits,

You've reached the end of the book with a working PDS. It signs commits,

Chapter 18 ended the main book with a working production PDS and a list of

A user's DID lives on a third-party log (plc.directory) and points at a

Chapter 13 handed clients a pair of HS256 JWTs whenever they typed the

So far in this book the PDS has been a server you talk to with curl. That's

The PDS is the only thing in the network that has a copy of every byte