beaker

tion, forking of sites, and early experiments with social applications such as Rotonde and Fritter. However, in his official post-mortem on the Beaker archive notice, Frazee outlined the practical limitations that led to discontinuation. https://github.com/beakerbrowser/beaker/blob/master/archive-... He explicitly noted that the project “never solved the hard problems” required for broad adoption, particularly for dynamic social networking. In a more detailed 2024 essay titled “Why isn’t Bluesky a peer-to-peer network?,” Frazee elaborated on the specific shortcomings of pure peer-to-peer models when applied to large-scale social systems. He concluded that insisting on a fully device-hosted peer-to-peer network for a mainstream social platform “would’ve been a mistake,” given users’ unwillingness to sacrifice features or reliability for theoretical decentralization benefits. The AT Protocol, which Frazee helped architect as Bluesky’s CTO (a role he continues to hold as of early 2026), represents a deliberate hybrid synthesis rather than an abandonment of prior principles. It retains core peer-to-peer innovations—cryptographically signed user data repositories, hosting agility, Merkle-tree-based verification, and portable identities—while delegating aggregation, indexing, and high-scale delivery to dedicated infrastructure (Personal Data Servers, relays, and AppViews). This design enables the data sovereignty and forkability that Dat/Beaker championed, while delivering the performance, discoverability, and moderation capabilities necessary for widespread use. The ongoing FreeSky initiative, discussed in our prior exchange, further advances this by providing independent Personal Data Servers and relays, reducing reliance on Bluesky-operated infrastructure and realizing more of the original portability vision. The Dat/Hypercore protocol itself was not discontinued; it continues under the Holepunchto organization and powers other applications. Thus, the technical lineage persists in parallel. In technology development, particularly within decentralized systems, iterative refinement based on empirical constraints is common. Frazee has publicly framed the transition as an application of lessons from multiple prior projects (including Secure Scuttlebutt and CTZN) rather than a repudiation. Whether one regards the shift as a misstep or a pragmatic evolution depends on the relative weighting of ideological purity versus practical adoption and usability at scale. Bluesky’s growth to millions of users and the expanding AT Protocol ecosystem suggest the hybrid model has achieved broader traction than pure peer-to-peer social experiments previously attained. In summary, FreeSky embodies the practical "alternative" envisioned in early AT Protocol discussions—offering decentralized hosting and tools within the Bluesky-compatible network rather than a separate platform. For those interested in trying it, start by exploring custom handles through freesky.social or reviewing the dashboard for operational insights. Additional details are available via Project Liberty announcements and AT Protocol documentation at atproto.com

plains about being hard to google, despite "AT protocol" and "Matrix protocol" or "Matrix.org" being trivial to search for :P There are some significant differences too: * Matrix aspires to be the secure communication layer for the open web. * AT aspires (i think) to be an open decentralised social networking protocol for the internet. * AT has portable identity by default. We've been working on this on Matrix (e.g. MSC1228 - https://github.com/matrix-org/matrix-spec-proposals/pull/122... and MSC2787 - https://github.com/matrix-org/matrix-spec-proposals/blob/nei...) and have a new MSC (and implementation on Dendrite) in progress right now which combines the best bits of MSC1228 & MSC2787 into something concrete, at last. In fact the proto-MSC is due to emerge today. EDIT: and here it is: https://github.com/matrix-org/matrix-spec-proposals/blob/keg... * AT is proposing a asymmetrical federation architecture where user data is stored on Personal Data Servers (PDS), but indexing/fan-out/etc is done by Big Graph Servers (BGS). Matrix is symmetrical and by default federates full-mesh between all servers participating in a conversation, which on one hand is arguably better from a self-sovereignty and resilience perspective - but empirically has created headaches where an underpowered server joins some massive public chatroom and then melts. Matrix has improved this by steady optimisation of both protocol and implementation (i.e. adding lazy loading everywhere - e.g. https://matrix-org.github.io/synapse/latest/development/syna...), but formalising an asymmetrical architecture is an interesting different approach :) * AT is (today) focused on for public conversations (e.g. prioritising big-world search and indexing etc), whereas Matrix focuses both on private and public communication - whether that's public chatrooms with 100K users over 10K servers, or private encrypted group conversations. For instance, one of Matrix's big novelties is decentralised access control without finality (https://matrix.org/blog/2020/06/16/matrix-decomposition-an-i...) in order to enforce access control for private conversations. * Matrix also provides end-to-end encryption for private conversations by default, today via Double Ratchet (Olm/Megolm) and in the nearish future MLS (https://arewemlsyet.com). We're also starting to work on post quantum crypto. * Matrix is obviously ~7 years older, and has many more use cases fleshed out - whether that's native VoIP/Video a la Element Call (https://element.io/blog/introducing-native-matrix-voip-with-...) or virtual worlds like Third Room (https://thirdroom.io) or shared whiteboarding (https://github.com/toger5/TheBoard) etc. * AT's lexicon approach looks to be a more modular to extend the protocol than Matrix's extensible event schemas - in that AT lexicons include both RPC definitions as well as the schemas for the underlying datatypes, whereas in Matrix the OpenAPI evolves separately to the message schemas. * AT uses IPLD; Matrix uses Canonical JSON (for now) * Matrix is perhaps more sophisticated on auth, in that we're switching to OpenID Connect for all authentication (and so get things like passkeys and MFA for free): https://areweoidcyet.com * Matrix has an open governance model with >50% of spec proposals coming from the wider community these days: https://spec.matrix.org/proposals * AT has done a much better job of getting mainstream uptake so far, perhaps thanks to building a flagship app from day one (before even finishing or opening up the protocol) - whereas Element coming relatively late to the picture has meant that Element development has been constantly slowed by dealing with existing protocol considerations (and even then we've had constant complaints about Element being too influential in driving Matrix development). * AT backs up all your personal data on your client (space allowing), to aid portability, whereas Matrix is typically thin-client. * Architecturally, Matrix is increasingly experimenting with a hybrid P2P model (https://arewep2pyet.com) as our long-term solution - which effectively would end up with all your data being synced to your client. I'd assume bluesky is consciously avoiding P2P having been overextended on previous adventures with DAT/hypercore: https://github.com/beakerbrowser/beaker/blob/master/archive-.... Whereas we're playing the long game to slowly converge on P2P, even if that means building our own overlay networks etc: https://github.com/matrix-org/pinecone I'm sure there are a bunch of other differences, but these are the ones which pop to the top of my head, plus I'm far from an expert in AT protocol. It's worth noting that in the early days of bluesky, the Matrix team built out Cerulean (https://matrix.org/blog/2020/12/18/introducing-cerulean) as a demonstration to the bluesky team of how you could build big-world microblogging on top of Matrix, and that Matrix is not just for chat. We demoed it to Jack and Parag, but they opted to fund something entirely new in the form of AT proto. I'm guessing that the factors that went into this were: a) wanting to be able to optimise the architecture purely for social networking (although it's ironic that ATproto has ended up pretty generic too, similar to Matrix), b) wanting to be able to control the strategy and not have to follow Matrix's open governance model, c) wanting to create something new :) From the Matrix side; we keep in touch with the bluesky team and wish them the best, and it's super depressing to see folks from ActivityPub and Nostr throwing their toys in this manner. It reminds me of the unpleasant behaviour we see from certain XMPP folks who resent the existence of Matrix (e.g. https://news.ycombinator.com/item?id=35874291). The reality is that the 'enemy' here, if anyone, are the centralised communication/social platforms - not other decentralisation projects. And even the centralised platforms have the option of seeing the light and becoming decentralised one day if we play our parts well. What would be really cool, from my perspective, would be if Matrix ended up being able to help out with the private communication use cases for AT proto - as we obviously have a tonne of prior art now for efficient & audited E2EE private comms and decentralised access control. Moreover, I /think/ the lexicon approach in AT proto could let Matrix itself be expressed as an AT proto lexicon - providing interop with existing Matrix rooms (at least semantically), and supporting existing Matrix clients/SDKs, while using AT proto's ID model and storing data in PDSes etc. Coincidentally, this matches work we've been doing on the Matrix side as part of the MIMI IETF working group to figure out how to layer Matrix on top of other existing protocols: e.g. https://datatracker.ietf.org/doc/draft-ralston-mimi-matrix-t... and https://datatracker.ietf.org/doc/draft-ralston-mimi-matrix-m... - and if I had infinite time right now I'd certainly be trying to map Matrix's CS & SS APIs onto an AT proto lexicon to see what it looks like. TL;DR: I think AT proto is cool, and I wish that open projects saw each other as fellow travellers rather than competitors.

OpenCollective and Patreon: 54%. You get the idea. Notably absent is "usage and retention went through the roof: 80%" and then "usage continued to grow for years: 100%." Everything that pfrazee wrote here about Beaker Browser at https://github.com/beakerbrowser/beaker/blob/master/archive-... is true for ssb.