AI Agent / Web3·WeaveVM (Load Network) · Arweave
WeaveVM (Load Network)
Hyperscalable EVM L1 settled on Arweave for permanent, low-cost storage. Solidity precompiles let contracts read/write Arweave data — purpose-built for AI datasets, model weights, and agent memory. Now branded Load Network.
- 01permanent on-chain AI datasets
- 02model weight storage
- 03agent long-term memory
- 04data-availability for L2s
- 05large-blob EVM apps
- pnpm add viem @weavevm/wvm-sdk
| Variable | Scope | Description |
|---|---|---|
| WVM_RPC_URL | Server | WeaveVM RPC endpoint (e.g. https://testnet-rpc.wvm.dev). |
| WVM_PRIVATE_KEY | Server | Private key for signing WeaveVM transactions and Arweave bundle uploads. |
Use WeaveVM to persist AI data on Arweave from Solidity. Configure a viem client with `chainId` = WeaveVM Alphanet/testnet and `transport: http(WVM_RPC_URL)`. Inside a contract, call the `arweave_read(txid)` precompile (address `0x17`) to fetch Arweave bytes, or emit a calldata blob that the wvm-archiver ETL pipeline pins to Arweave automatically. From JS, read pinned data with the wvm-data-retriever (`wvm://{txid}`). For agents: store model checkpoints, embeddings, and conversation logs as Arweave blobs and reference their `txid` from on-chain state.
- ⚑Arweave finality is delayed — data is queryable from the WeaveVM gateway in seconds but globally permanent only after ~30 minutes of confirmations; do not assume immediate cross-gateway availability.
- ⚑Precompile gas pricing for Arweave reads is non-trivial; cache `txid -> bytes32 hash` mappings on-chain rather than re-reading large blobs.
- ⚑Storage is permanent and immutable — there is no delete; never write secrets, PII, or unaudited model weights you might want to revoke.
- ⚑The project rebranded from WeaveVM to Load Network in 2026; pin SDK versions and watch for package renames in `@weavevm/*` -> `@loadnetwork/*`.
- ⚑Network reliability for the bundler/archiver depends on Arweave miners; build retries and fallback gateways into upload paths.