Amazon Web Services (AWS) and the University of California, Berkeley have introduced AWS Infinity, an open-source runtime for artificial intelligence agents that changes how these systems wait for answers. Instead of keeping a process running while a tool does its job, Infinity suspends the agent, saves the conversation history and resumes it when the result arrives.
The project is part of Hydro, an AWS initiative with UC Berkeley, and is distributed under the Apache 2.0 license, according to Dosa AI Tools' summary of the announcement. The core idea is to treat a tool call as a suspension point, similar to stackless coroutines in programming languages.
In practice, this targets a known agent bottleneck: much of the time is spent waiting. A coding agent may start a build that takes twenty minutes, a triage agent may wait hours for a new pull request, and an assistant may wait until the next morning for an approval. Keeping thousands of such agents awake, each holding a process, costs memory and compute even when nothing is happening.
The headline number: 75,000 agents in 8 GB
According to AWS, an idle agent in Infinity takes about 100 KB. That would allow around 75,000 concurrent agents to fit in the 8 GB of memory of a Raspberry Pi 5. The figure is AWS's own illustration of memory savings, not a performance guarantee for every workload: real consumption depends on the model, context and tools used.
The runtime is also designed for serverless environments. AWS cites use with AWS Lambda, Amazon SQS and Aurora DSQL, so no compute needs to stay active while agents wait. Asynchronous events arrive through synthetic tool calls, within model API constraints.
Agents that talk to each other
Another feature is built-in tooling for agent-to-agent communication: spawn_thread and report_to_parent. Events are processed in child threads by default, to keep the parent agent's context from growing unchecked with every new event.
For developers, the appeal is direct: long-running, event-driven agents become feasible without a fleet of always-on servers, both in local processes and on serverless platforms. For companies, the potential gain is the cost of keeping large fleets of support, triage, automation and coding agents on standby.
Like any claim about scale, Infinity still needs to prove itself beyond AWS's examples: independent benchmarks, behaviour under failure, resume latency and the security of agents suspended for long periods are points to watch in future releases and community adoption.
Image credit: Amazon Web Services logo, via Wikimedia Commons.
Sources: AWS Open Source Blog, via AWS News Feed, "Introducing Infinity: a runtime for extreme agent concurrency" (October 23, 2026); Dosa AI Tools, "Moonshot's Kimi K3.1 Pro lands, open weights due Nov 4" (October 24, 2026); Hydro / Infinity project repository on GitHub, indicated in a public Instagram post; public Instagram post by @ai.dailypulse about Infinity (October 24, 2026).
Transparency: this article was produced with the assistance of artificial intelligence, cross-checking public social media posts and internet sources. The scale figures cited are AWS disclosures, not independent test results. Consult the original sources to confirm the details.
By Muse
