Blog
Latest Articles
Product updates, engineering notes, and governance stories from the SOIT team.
Our agent loop is 239 lines. Frameworks and runtimes are not the same layer.
A framework decides how an agent thinks; a runtime decides whether the moment it reaches out counts. Counting our own repository line by line shows where that boundary sits, and which tool keeps it there.
Read moreYour agent request touches twelve containers. Only one of them runs a model.
A walk through the SOIT quickstart topology, service by service: what each container is responsible for during a single run, and which guarantee disappears if you remove it.
Read moreI tried to cut our twelve-container stack down to four. Two of my three conclusions were wrong.
How far the SOIT quickstart topology can actually be trimmed for a demo, which service you fold in rather than delete, and the two things reading the code got wrong that only running it revealed.
Read moreFive questions to answer before you put MCP in production
MCP standardised how an agent gets tools. It says nothing about who may call one, where the credentials live, what the call can reach, what evidence it leaves, or whether you can replay it.
Read moreSOIT is now open source: a governed runtime for AI agents
SOIT is open source and v1.0.0 is released: an agent runtime where permissions, secrets, egress, audit, cost, trace, and replay are kernel concerns.
Read moreIntroducing SOIT
Introducing SOIT, a governed agent runtime for enterprise AI systems.
Read more