Built by one person, not a
funded team.
That's worth knowing upfront, not hiding behind a fake "team" page — it shapes what Agent Substrate is honest about being ready for, and why self-hosting exists as a real, first-class option.
Why this exists
Most AI agent pilots never make it to production — not because the models aren't capable enough, but because nobody can see what an agent actually did, or stop it before it does something wrong. Governance gets bolted on after the fact, if at all.
Agent Substrate is built the other way: durable execution, human approval gates, and full tracing aren't optional add-ons, they're how the runtime works. A run can pause for a human, a schedule, or a webhook without holding compute, and resume with a complete record of what happened before it stopped.
The engine and chat UI are open source — you can read exactly how the durability and guardrails work, or run the whole thing yourself with no dependency on this platform at all.