🥷AI Agent Contract

This section will dive into the technical specifications of the AI Agent Contract's execution. All content will focus on the off-chain components that allow for developers to host their Agent scripts or general purpose JS programs on Phala Network as a serverless backend.

Phala's WASM Runtime (Wapo)

The AI Agent Contract is served through Phala's gateway (https://wapo-testnet.phala.network/ipfs/<cid>) and executes in a TEE compute node on Phala Network. Phala enables the execution of JS programs through a custom built WASM runtime called Wapo.

Wapo is a poll-based async WebAssembly runtime that provides extended networking support for guest WASM applications. It originated as a module called SideVM in Phala PRuntime, now independent, and the execution engine has been switched from wasmer to wasmtime.

Why Wapo?

Wapo is mainly designed as the next version of the AI Agent Contract execution engine. However, it can be used as a general-purpose WebAssembly runtime with networking support. The AI Agent Contract execution engine today is based on Substrate's pallet-contracts, which is not flexible enough and also has many limitations. Here is the table of comparison between 2.x and 3.0:

	AI Agent Contract 2.0	AI Agent Contract 3.0
Program Type	WebAssembly	WebAssembly
VM Engine	wasmi	wasmtime (faster)
Incoming Networking	Query RPC	Query RPC or Listening on TCP port
Outgoing Networking	HTTP requests (with time and size limit)	Arbitrary TCP connections
Execution mode	Transaction/query-based, 10s limit	Long-running or query-based, no tx
App Memory	4MB for ink / 16 MB for JS	Up to 4GB due to wasm32 limit

With the enhancements with Wapo, Phala has taken the next steps to port the QuickJS implementation to execute in Wapo runtime called WapoJS.

WapoJS

WapoJS is an upgrade from the previous JS runtime called SideVM QuickJS. With this upgrade, developers can push the boundaries with their general-purpose JS programs with a faster response time vs the previous implementation.

Feature	SideVM QuickJS	WapoJS
VM Memory	16MB	100MB (default) Up to 4GB
Maximum Execution Time	60 seconds (default) to unlimited	60 seconds (default) to unlimited
HTTP Request API	Asynchronous fetch API	Asynchronous fetch API
Websockets	Not Supported	Supported (not enabled, yet)
Concurrent Requests	Fully supported	Fully supported
Execution Speed	Fast	Faster
SCALE codec API	Sidevm.SCALE	Wapo.SCALE
Max Code Size	500KB	Unlimited (Could be limited by streaming receiving from fetch + AbortController)
Limited CPU Burst	CPU time between async calls is limited. e.g. Too complex for-loop may hit the burst limit.	CPU time between async calls is limited. e.g. Too complex for-loop may hit the burst limit.

For more information or inquries about increasing some resource limits, reach out the Phala Network team on discord.