NOVAInetwork

AI-native L1 blockchain for agent payment
settlement. Built from scratch in Rust. 90K+ lines,
1,500+ tests, open source.

[NOVAInetwork @NOVAInetwork]

Two trends are converging that most people are
treating as separate.

Real world assets are moving on-chain. AI agents
are becoming economic actors. McKinsey projects
$3-5T in agentic commerce by 2030.

Both need infrastructure that does not exist yet.

I have been building it for six months. Here is
why.

https://t.co/gILLaTMUMo

[NOVAInetwork @NOVAInetwork]

Interesting that the founding set is the incumbent rails plus the hyperscalers. That tells you which problem x402 is actually solving: making human-owned payment instruments legible to agents. Useful, and probably the right first step. The part it leaves open is what settles when the agent is the principal rather than the proxy, no card behind it, paying for its own compute out of what it earned. Does x402 have a story for that, or is it deliberately scoped to the delegated case?

[NOVAInetwork @NOVAInetwork]

writing about this because I'm looking for a technical co-founder for NOVAI. AI, Rust, distributed systems, consensus.

the code is the pitch: https://t.co/t79jxOPA88

full writeup: https://t.co/Sbor9JzKNw

[NOVAInetwork @NOVAInetwork]

I build an L1 blockchain alone. No team to argue with, no reviewer to catch the idea I'm in love with that's actually bad.

you don't get more careful by trying harder. trying harder comes from the same judgment that made the mistake.
so I don't trust myself. I run two processes instead.

[NOVAInetwork @NOVAInetwork]

the real reason: every batch ships on top of the last one. I want the next set of code landing on a strong base, not a codebase full of holes that's one incident from collapse.

a solo founder needs this more than a team, not less. it's the second set of eyes I don't have.

[NOVAInetwork @NOVAInetwork]

The "not from AI" part is the one I keep turning over, because I am building an AI-native L1 and I think you are partly right. The wave does not come from bolting AI onto crypto as a narrative. Most of that is corporate theater. Where I think it does come from is the unglamorous version: agents needing rails they cannot get elsewhere, identity, verifiable execution, settlement that does not require trusting one provider. That is infrastructure, not an AI story. If the next wave is anti-corpo and anti-narrative like you are betting, the AI angle that survives is the one that looks like plumbing, not the one that looks like a pitch deck.

[NOVAInetwork @NOVAInetwork]

Agree on both, with one caveat from building an L1 right now: post-quantum is a swap-the-primitives problem you can largely plan for, but privacy changes your whole execution and state model, so it has to be a day-one design decision, not a later bolt-on. Retrofitting privacy onto a transparent chain is the hard part most people underestimate.

[NOVAInetwork @NOVAInetwork]

Overnight soak result on NOVAI: 4 validators in lockstep, ~230k blocks, identical state root the whole way, zero errors.

this is after deploying a consensus safety fix I traced from a live testnet incident this week. found it, proved it, fixed it, shipped it.

solo, an AI-native L1, building in public. the foundation is getting solid. deeper writeup next week.

[NOVAInetwork @NOVAInetwork]

Then, same night, a real one: a node committed a divergent block and got stuck rejecting the canonical chain. Caught it, stopped the node, chain stayed healthy on the other 3.

Building an L1 solo means you see every edge case yourself. This is the work.

[NOVAInetwork @NOVAInetwork]

Shipped persistent vote tracking on my L1 testnet today.

The fix: a validator now writes its vote to disk (force-fsynced) before broadcasting it. On restart it refuses to vote again at any view it already voted at.

Closes a restart-race where a node could equivocate after a crash.

[NOVAInetwork @NOVAInetwork]

The subtle part: I keyed the durable record by (height, round), not height alone. A height-only guard wrongly blocks the legitimate re-proposal after a view change and halts the chain. I learned that one the hard way already.

[NOVAInetwork @NOVAInetwork]

deployed a consensus safety fix to the live testnet two nights ago. found a bug where one byzantine validator could drive two honest nodes to commit conflicting blocks at the same height. fixed it, proved it with a regression test, shipped it.

soaked overnight: 200k+ blocks, all 4 validators in agreement, zero forks, zero view changes. solo.

[NOVAInetwork @NOVAInetwork]

My testnet wedged at a single block height for a week and would not recover. Four validators, zero restarts, stuck.

The lesson that stuck: the loudest error in the logs was not the real bug. The actual fault was quieter, underneath it, a commit cursor that had silently outrun the certificates meant to justify it. Fixing only the visible symptom would have left the chain just as stuck.

Wrote up the whole thing, three layers deep, plus the four-part fix and an honest section on what is still open:

https://t.co/GaX54LJrQR

[NOVAInetwork @NOVAInetwork]

MCP over chain data is the right direction but Dune-as-intermediary is a halfway step. The interesting version is agents reading settlement state directly off the chain rather than through an analytics layer that can lag or reindex. Stablecoin flow is exactly the data agents will want firsthand. Natural-language-over-Dune is the demo, agent-native data access is the product.

[NOVAInetwork @NOVAInetwork]

Deposit-from-35-chains is the underrated half. Cross-chain asset abstraction at entry is what makes the leverage usable at all. Confidential perps is the real test though. Hiding positions without breaking liquidation determinism is the hard part. If liquidators can't see what to liquidate, you either leak through the oracle/keeper path or you push it to a privileged actor. Which way are you leaning?

[NOVAInetwork @NOVAInetwork]

@RoundtableSpace Determinism harness for a BFT L1. 4-validator in-process test that catches state divergence on any non-merkle-authenticated state field. Built it yesterday after a single-validator fork incident reminded me why "all 4 nodes agree" is a stronger gate than "1924 unit tests pass."

[NOVAInetwork @NOVAInetwork]

Hardened and reliable mostly means "has survived adversarial load." Most L1s in the top 50 have never been stress-tested by autonomous high-frequency agent traffic, just by human-paced trading. The "reliable" set after AI agents become real users is probably smaller than people think.