Hyperspace is the most used AI application where every “user” is a cryptographic wallet signing every action. However in order to record such actions on a blockchain, the current cost and UX trade-offs are challenging to navigate. It’s cheaper for us to build our own ledger.

toly 🇺🇸 Dankrad Feist mteam.eth 🗼 BREAD | ∑: ZKP recursive batching is not the bottleneck, the issue is all about state. This has been the outline of our design - where based on knowing which transactions impact which state, we “send” them to relevant state committees. What does this yield (per theory): 200k TPS with 4.4s

toly 🇺🇸 Dankrad Feist mteam.eth 🗼 BREAD | ∑: Here's how it scales in our design, which involves generating STARKs based on execution traces (WASM), and then batching STARKs into SNARKs using Plonky2 and "simple batching" instead of other batching methods as this figure shows. docs.google.com/spreadsheets/d…

The hard thing about “the man in the arena”. ———— Kudos to Varun for being one. ———— In the era of exponential tech, most things either totally die and some rare few things grow exponentially, there is simply no “good enough” (even great is insufficient). This means the

Varun Dankrad Feist mteam.eth 🗼 BREAD | ∑: Happy to jump on a call. Async execution would allow for very big batching and can be as far as 1 day behind the tip.

Even inside crypto there is massive adverse selection. The teams that care about building agentic workflows, ux and products like Hyperspace are out there building open source ai agents with p2p networks to serve them. Try this out: compute.hyper.space

What Decentralized AI looks like 💻 10,000+ connected nodes 🌏 1,000+ unique cities 🏴‍☠️ 10 million+ liveness checks 🤖 500+ open source models In 2009, you could run a simple piece of software on your machine and participate in a new type of a peer-to-peer system called Bitcoin.

“..It will be a normalized world for your AI startup to also have a crypto strategy (just like we are doing at Hyperspace + Hyperspace Protocol)”

What’s in a state ? At Hyperspace Protocol we are building our own agentic blockchain protocol for one simple reason: state. A world of billions of AI agents with their own state is simply not feasible on any modern chain, so we must stretch it further and invent the future. When

The most decentralized AI network

We would be rolling out on-chain, verifiable inference for this entire network within a few days. One year ago we wrote a paper called “BitTorrent for AI”, and since then have learnt a lot experimenting and building real products and networks.

🚧 We are actively working on resolving an issue with the points system 🚧 Meanwhile, you can read about how the system works over here: flops.hyper.space and start running a node at node.hyper.space (recommend: download a client, all major OS supported)

we are building a blockchain natively for AI agents with these features: - high throughput due to massive parallelization - low cost payments routed by stake-incentivized nodes - state model which is designed for peak concurrency (think ATMs)

agentic browser, by hyperspace tl;dr: answer engines are a thin-wrapper around search results and LLM APIs which don’t have data network effects; at Hyperspace we are innovating upon a new software-type: the “agentic browser”, which pokes at the idea of how do the agents rank

We are building the fastest blockchain for AI agents to transact Hyperspace Protocol. At 200k TPS with a 4.4 second finality, this design scales.

🎄🎅 announcing the Hyperspace Wallet: the intelligent agent for interacting with blockchains

The accelerationism here is out of control. Kudos Varun

we will network all of them more powerful x cheaper x abundant NVIDIA computers x more high-quality data x more local models -> more INTELLIGENCE in the network these are over 20,000 nodes connected right now to one P2P network, weirdly, more than all the blockchains combined

Hyperspace is a new kind of network that routes vectors, models, agents, compute, tools and payments — across a global peer-to-peer layer. It operates 99% tokenless, enabling open access to routing and discovery using mechanisms similar to DHTs and BitTorrent. For priority