Hi, I’m the founder of AssetCLI, the team that just won Solana’s AI-Agent MCP hackathon for re-imagining token launches.

On-chain capital markets are exploding. Hyperliquid’s purpose-built exchange chain cleared US $248 billion in perpetual-futures volume last month, a 51 % jump in thirty days (tokeninsight.com, x.com). Meanwhile Pump․fun’s memecoin factory is racing toward a billion-dollar annual run-rate as Solana’s top-grossing app (wired.com, news.bitcoin.com). Yet the first block of most launches is still hijacked by Telegram snipers, who vacuum up 80 % of supply and dump on real fans (blockworks.co, x.com).

Why does this happen? Creators must hand over control to generic launchpads, and changing a curve still means writing low-level Solana programs—something large-language models compile correctly less than a third of the time on public benchmarks (arxiv.org).

AssetCLI fixes that today. Our live agent turns a plain-English prompt—“raise a million dollars on a linear curve” —into production-ready Solana code: mint, curve maths, automatic Raydium pool.

Tomorrow we go beyond code into vertical control just like Hyperliquid. Hyperliquid hard-wires an order-book into its own chain to give traders deterministic sequencing and zero gas. AssetCLI does the same for token launchpads, but we hard-wire the code-gen pipeline itself

1. CurveScript. A 30-keyword DSL with no syscalls or signer arrays, shrinking the LLM search space by ~100× and scoring > 95 % first-compile success in our pilots

2. CurveGPT. The agent that writes only CurveScript from plain English prompts like “Raise $1 million on a linear curve that auto-migrates at 300 SOL reserve.

3. Compiler + Proof Engine. Converts that script into WASM and proof of overflow-free maths and non-negative reserves before we let it anywhere near main-net .

4. CurveVM. A micro-VM inside our roll-up that runs only four op-codes—buy, sell, add_liquidity, migrate_to_amm—so every call stays below 300 k compute units versus Solana’s 1.4 M ceiling and executes in parallel with zero account collisions

• AssetL2 roll-up. Anchored to Solana but sequenced by AssetSequencer-BFT - a Hyperliquid style consensus that ultimately settles to Solana every second and gives you bot-proof ordering of transactions.

from src.curvescript import parse
from src.compiler import compile_program
from src.curvevm import CurveVM
from src.rollup import BatchPoster, FakeSolanaClient

script = """BUY 5\nSELL 2\nADD_LIQUIDITY 3\nMIGRATE_TO_AMM 1"""
program = compile_program(parse(script))
vm = CurveVM()
vm.execute(program)
client = FakeSolanaClient()
BatchPoster(client).commit(program)

Hyperliquid’s edge is owning the trading state machine; our edge is owning the code-generation state machine. Creators get deterministic first-block fairness, gas-free UX, and a cryptographic proof that the AI-written curve can’t rug pull - these are advantages that generic pads and generic code-gen simply can’t match.

If you believe the next Nasdaq will be described in chat, not coded in Rust, join us. AssetCLI lets you speak your raise today and will guarantee it’s safe tomorrow. Thank you.

Local testnet

The testnet crate provides a lightweight demonstration of the migration plan’s step 5. It mints $ASSET using assetvm, posts a batch via the sequencer and prints the resulting signature. Run it with:

cargo run -p testnet

Sharded Perp DEX Integration

We have adapted the sharded perp exchange design to run through CurveScript, CurveGPT, the compiler+proof engine, CurveVM, and the AssetL2 rollup. See SHARDED_PERP_CURVE_INTEGRATION.md for the implementation blueprint that keeps all router/slab invariants while highlighting AssetL2’s unique pipeline.