By Roman Aglaganov
AI agents spend much of their time waiting. A coding agent may start a build that takes twenty minutes; an assistant may wait until morning for someone’s approval. In most agent software, the agent holds on to a running process and its memory the whole time. On 23 October AWS released Infinity, an open-source runtime built with UC Berkeley’s Hydro project, to change that.
How it works
Infinity borrows an idea from programming languages, which pause a function while it waits for a network request and resume it later. When an agent calls a tool, Infinity sends the request, saves the conversation and suspends the agent. When the result arrives, any machine with access to the saved conversation can pick the agent up again. Because the conversation holds everything needed to continue, a waiting agent uses no memory beyond its stored history.
The numbers
In AWS’s benchmark, idle agents took about 100 KB each, so roughly 75,000 would fit in the 8 GB of a Raspberry Pi 5. The test used a simulated model and in-memory storage, so it measures the runtime’s own overhead, not real-world speed. Infinity also works on serverless platforms such as AWS Lambda: the function can end while a tool runs, even for days, and nothing runs in the meantime. AWS’s reference setup uses its own queue, database and scheduling services.
“Infinity is an early research release, and we are sharing it to get early feedback,” AWS wrote. It is written in Rust and released under the Apache 2.0 licence.
Why it matters
If companies run thousands of agents that mostly wait on builds, approvals and other people, keeping each one alive gets expensive. Treating an agent as saved data that wakes up on demand is a simple idea that could cut those costs sharply. It also suits AWS, whose cloud services would be the obvious place to run it.
