---
id: PROV.1
status: ratified
ratified: 2026-09-01 (Jason Woltje; PRD rev1 ratification PR)
---

# PROV.1 — Provider configuration

A provider is a model-inference source: hosted (Claude, OpenAI, ZAI, N others)
or local (Ollama, LM Studio, other).

## Provider configuration surface (WebUI page + CLI)

| Control | Notes |
|---|---|
| provider selection | dropdown of supported providers |
| name | user-chosen instance name (multiple named instances of one provider type are expected) |
| auth mode | OAuth or API key — the account itself lives in [[AUTHN.1-auth-accounts]] |
| local provider setup | endpoint/port for Ollama, LM Studio, other local providers |
| activate / deactivate | inactive providers are not selectable downstream |

Provider records are runtime state (Postgres-owned, projected). Credentials
never enter provider records; they live with the credential broker
([[AUTHN.1-auth-accounts]]).
## E2 inputs (triage 2026-08-31; see [[GOV.2-docs-inventory]])

**Canonical ground truth**: `DEVELOPER-GUIDE/architecture/decisions/mos-runtime-portability-m1.md`
(the only current ADR for the logical-agent/connector-lease identity model);
`ADMIN-GUIDE/operations/mos-connector-lease-operations.md` — connector
activation is a **deliberate deny-all hold**; nothing in this section may imply
it is live.
**Drafts noted**: `rfcs/optional-ai-egress-gateways.md` (non-operative;
separates `IProviderAdapter` from egress-gateway concerns).