Tempest

Tempest is a self-hostable AT Protocol Personal Data Server built with Elixir and Phoenix. It implements the account, identity, repo, sync, blob, and admin surfaces needed to run and inspect a small PDS.

Why I Built This

I wanted a PDS implementation in a functional language that I understood from top to bottom: accounts, handles, DIDs, record writes, signed commits, repo exports, blob storage, and the XRPC layer. Elixir fits the shape of the work because Phoenix gives the server a solid HTTP foundation while OTP keeps the long-running pieces explicit.

Features

• XRPC routing for AT Protocol server, identity, repo, sync, blob, and admin endpoints
• Account creation, password hashing, sessions, refresh tokens, and bearer auth
• DID document generation, hosted DID discovery, handle validation, and handle resolution
• Per-account repository storage with blocks, records, commits, and repo metadata
• Record writes through createRecord, putRecord, and deleteRecord
• Record reads through getRecord, listRecords, and describeRepo
• Sync reads through getRepo, getLatestCommit, getRecord, getBlocks, getRepoStatus, listRepos, and listBlobs
• CAR v1 responses for repository export
• Local blob storage with upload, listing, and account UI inspection
• Operator pages for account, repo, blobs, access, security, migration, sequencer, and firehose views
• Admin status and compatibility pages guarded by bearer auth
• Protocol-shaped JSON errors, verb mismatch handling, and validation boundaries

Architecture

Tempest uses Phoenix and Bandit for HTTP, Ecto with SQLite for account data, and raw per-DID SQLite databases for repository storage. The repository core includes pure-Elixir implementations for AT URIs, NSIDs, record keys, DIDs, TIDs, CIDs, DAG-CBOR, CAR v1, Merkle Search Tree operations, and signed commit blocks.

Every record write runs inside a SQLite transaction that updates MST entries, writes CBOR blocks, signs a new commit, updates record indexes, and advances repo metadata.

Status

Tempest is early pre-release software. The current work is focused on durable firehose sequencing, blob garbage collection, broader lexicon validation, OAuth, app passwords, admin tooling, repo import/export, and compatibility testing against official fixtures and SDKs.