The core idea behind Gace AI was to treat AI plugins and tools like standard JS functions or packages, rather than the more heavyweight approach seen in MCP which requires a persistent server.
The biggest challenge was achieving this without the typical latency associated with spinning up new sandboxes. We eventually landed on using quickjs-emscripten, which utilizes a WASM-compiled QuickJS engine to execute JS scripts in a secure, isolated environment.
By going this route, we managed to achieve 0.025s cold starts for new sandboxes. This allows us to offer a free hosting and a hot reload experience where you can host and call tools instantly without worrying about the infrastructure overhead.
The core idea behind Gace AI was to treat AI plugins and tools like standard JS functions or packages, rather than the more heavyweight approach seen in MCP which requires a persistent server.
The biggest challenge was achieving this without the typical latency associated with spinning up new sandboxes. We eventually landed on using quickjs-emscripten, which utilizes a WASM-compiled QuickJS engine to execute JS scripts in a secure, isolated environment.
By going this route, we managed to achieve 0.025s cold starts for new sandboxes. This allows us to offer a free hosting and a hot reload experience where you can host and call tools instantly without worrying about the infrastructure overhead.