most ledgers make AI agents wait in line.
a traditional chain advances on a physical clock — every transaction squeezed into one global total order, one block at a time. that's fine when humans click buttons a few times a minute. it falls apart when thousands of AI agents act concurrently, each doing work that has nothing to do with the others. forcing them into a single sequence is an artificial bottleneck: agents idle, waiting for a global tick that has no bearing on what they actually did.
setu throws out the global clock.
it's a causally-driven ledger for AI agents: it advances only when actual intelligent work occurs, and it orders that work with a vector logical clock (VLC) instead of wall-clock time. a VLC doesn't ask "what time is it?" — it asks "what did this depend on?" two agents doing unrelated work are recorded as exactly that: concurrent, with no false ordering imposed. work that genuinely depends on prior work keeps its causal order, precisely.
the result is a causal-DAG, not a single chain — a record of what happened and what it followed from. ordering by causality instead of by a global sequence is what lets the substrate run agents in parallel without the consensus overhead a total-order chain pays at every step. the design targets that fall out of this — 200K–300K TPS, 50–100ms finality, zero gas — are a consequence of not making unrelated work wait, not a number bolted on top.
and because the ledger is a record of work, the work has to be real. setu attests execution inside TEE enclaves: an agent's action is verified to have actually run, the way it claims, before it enters the ledger. ordering answers "in what order did this happen"; attestation answers "did it happen at all." you need both for a record that other agents — and their payments — can rely on.
most infrastructure for AI agents borrows a design built for human transactions and hopes it scales. concurrent, autonomous work needs a substrate that orders by causality and proves what was done. that's setu.
setu.hetu.org