A social graph is the map of who follows whom and who reacts to what. On a closed platform that map is the product, and it is the reason you cannot leave. In a decentralised design the map is shared infrastructure any client can read, which changes who has leverage — and it turns out that the hard problems move rather than disappear.
Three approaches, three different trade-offs
Lens issues a profile as an NFT and stores social actions as on-chain primitives, so a follow is a transaction the user owns. That gives the strongest possible guarantee of portability and the worst possible cost profile: every social action is a state change every node must process. Lens has since moved toward its own chain to escape that constraint, which is a sensible engineering choice and a reminder that "on-chain social" at Ethereum L1 prices was never viable at scale.
Farcaster takes the opposite bet on cost. Messages live in an off-chain peer-to-peer layer with a periodic on-chain commitment, while identity and key ownership are registered on-chain. A user's account is a set of keys tied to an on-chain registry, and their posts are signed messages that anyone can fetch and verify. The trade is latency and cost for a weaker guarantee: the graph is portable because it is signed and public, not because every post is a transaction.
Nostr goes furthest. There is no chain and no token, just signed events relayed by whoever chooses to relay them, with the protocol defined by a set of NIPs. It is the most censorship-resistant and the least enforceable: nothing stops a relay from dropping your events, and nothing stops you from moving to another relay.
What portability actually requires
Portability is the entire argument, and it depends on a standard to port into rather than on a database schema. Two questions separate the designs that work from the ones that do not.
Can a third party build a client tomorrow without asking permission? If reading the graph requires an API key issued by the incumbent client, the graph is not portable regardless of where it is stored. This is why interfaces matter more than storage: Farcaster's Frames specification and an open social API do more for portability than the underlying architecture.
Does the user keep their followers if they leave? If followers are addresses that followed a profile ID, then yes — the follow relationship is data, not a subscription, and a new client reads the same data. If followers were accounts inside an application database, then no.
Where the difficulty actually goes
Indexing. Reading an open graph efficiently is genuinely hard. A client cannot scan every event on every relay at every page load, so somebody builds an index — and the indexer becomes the new centralisation point. This is the quiet failure mode of decentralised social: the protocol is open and the practical read path runs through one service.
Moderation. Moderation does not disappear when the graph is open; it relocates. A relay can refuse to serve content, a client can refuse to render it, and an indexer can refuse to include it. The difference is that a user can choose a different client — but only if they know which one filters what, which is a worse user experience than a single platform making one consistent decision.
Cost of writes at scale. Anything that makes every post a transaction caps throughput at what the chain can handle and prices out normal usage. Commitment schemes and dedicated chains are the workarounds, and each one reintroduces some centralisation.
What to watch
The signal worth tracking is not user counts, which are heavily incentivised and easy to inflate, but whether third-party clients exist that the original team does not operate. A decentralised social network with exactly one viable client has the architecture of a decentralised network and the governance of a platform.
For a creator, the practical version of all this is simple: if your audience lives in a graph you can export, you have a hedge against any single platform's decisions. If it lives in a follower count on one app, you do not — and no amount of on-chain infrastructure changes that until you have actually exported it once and confirmed the audience came with you.
References
Next, read Farcaster Protocol 101. More in Web3 Applications.