Five tables, four kinds of replay, and a status called in_doubt
A source-level walkthrough of the run/step/tool-call ledger behind "replayable agent runs" — including the seven places it does not line up yet.
Sep 10, 202618 min read
Command Palette
Search for a command to run...
Latest articles
Agent frameworks and agent runtimes are not the same layer, and the code says where the line is
Why I ran wc -l on my own repo Someone asked again — roughly the tenth time since we open-sourced — whether SOIT is "just another LangChain". I have answered that with architecture diagrams before and
Sep 8, 202616 min readWhat are the other eleven containers for? Walking one agent run through the whole stack
You type docker compose up -d on this project's quickstart, and the terminal answers with a wall of container names — a dozen of them, before a single request has been served. That wall is where most
Sep 5, 202617 min readReading the code told me which services I could drop. Running it told me the truth.
I read our own source code to work out which services a demo could do without, wrote down three conclusions, and then ran the trimmed stack to check. Two of the three were wrong. That is the real subj
Aug 31, 202615 min readGoverning MCP tool calls: permissions, secrets, and egress
Wiring up an MCP server is an afternoon of work. One streamable HTTP endpoint, one list_tools call, and the tools land in the model's callable list. That part is genuinely solved. Everything after it
Aug 29, 20269 min readEvery agent run is a governed run: the architecture behind our open-source agent runtime
We open-sourced SOIT a few weeks ago. The one line version: an agent runtime and governance platform for teams that need AI agents to touch real enterprise systems without losing control. This post is
Aug 27, 20266 min read