Slot Gateway.
Sonata42 is an AI research company. Slot Gateway is one of its projects. It aims to give AI agents that work at the same time fair and orderly access to the tools and resources they share.
The problem.
More and more software is built from many AI agents that run at the same time. These agents often need the same tools and resources, such as a code repository, a database, a browser session, a device, or a paid service with a rate limit. Many of these tools can only be used safely by one agent at a time.
When several agents reach for the same tool at once, their actions can collide. One agent may overwrite the work of another, a shared resource may be left in a broken state, or an agent may wait without end while others take every turn. Common answers today are locks, queues, and retries added case by case. These do not give a clear rule for who goes next or a clear limit on how long anyone waits.
The aim.
Slot Gateway is planned as an open source gateway, released under the Apache 2.0 licence, that sits between agents and the tools they share. The gateway gives each agent a time slot at a tool. During its slot, an agent has exclusive use of that tool. When the slot ends, the next agent takes its turn.
The goals are fair turns for every agent, exclusive access during each turn, and behaviour that is predictable and easy to inspect. The gateway is meant to work with existing agent frameworks and tool protocols, so that teams can adopt it without rewriting their agents.
Status.
The project is at an early stage. The underlying protocol is described in a public preprint, available on Zenodo at doi.org/10.5281/zenodo.23753357. The protocol has been tested with a working simulator, which will be published as open source code. The gateway software itself is the next stage of the project. It will also be released openly under the Apache 2.0 licence, and this page will link to both when they are available.
Who.
Slot Gateway is developed by the Sonata42 team. It sits within the company's wider research on the routing and protocol layer for AI systems, described on the research page.