What you deploy
You do not install “a chatbot”. You install a sovereign organizational unit: a department, a firm, a ministry, a team — complete in identity, memory, work queue, governance and the ability to federate.
The unit of delivery is agents-host
Each agents-host is a digital department. It has its own identity, its agents (each with an identity document, memory and tools), its local Task Queue and its boundary. You scale by adding hosts, not by inflating one.
The per-agent runtime is interchangeable. What SiS adds is the institutional layer around it: registry, social memory, judgement, human gates, provenance.
Three surfaces for three jobs
| Surface | Who uses it |
|---|---|
| Conversation UI | Whoever commissions the work (pipelines, institution interview, corpus query) |
| Console | Whoever administers: Operate (panels) and Explore (live graph) |
| Command line | Whoever installs and diagnoses |
On-prem, dedicated stack
Each customer gets their own installation. Data multi-tenancy is the last horizon on purpose: the on-prem product does not need it. [V]
What work produces
A commission walks through roles (researcher → writer → auditor), leaves a deliverable with a seal and sources, and is traced in the graph: who asked, who produced it, from which corpus, under which approval. [V]