Skip to content
LQ.AI

Hand over LQ.AI to a client

  • partner

Before handing over an installation, agree who will maintain it, manage access, and choose the AI services it can use — and check the license and branding requirements for the version you’re delivering. You are here because you run LQ.AI for someone else — a client, a portfolio company, an internal business unit that is not “you” — and you need to know what you can promise them about the deployment before you promise it. That is a narrower question than “how do I install it” (/operate/) or “is the data handling sound” (/trust/); this namespace is about the commercial and licensing surface of delivering a branded deployment.

What ships here today

Section titled “What ships here today”

Only one page: branding and licensing obligations. It answers the question that has an actual answer in the repository right now — what the upstream license lets you do with LQ.AI’s chrome, and where the line sits. It does not supply a complete production-readiness assessment, support contract, theming manual, or multi-client operating plan — use the installation, maintenance, and privacy guides to assemble the rest of the handover, and agree the remaining responsibilities with the client.

The rest of this namespace — a stitched deployment guide across /operate/, /skills/ and /trust/; the theming mechanics (which CSS variables a branded deployment actually sets); multi-deployment operations; a production-readiness matrix; what you may and may not claim to a client; a hand-off pack for a client’s security team — does not exist yet. The docs-site mini-PRD (PR #511) leaves the framing for this namespace as an open question (its open question 4) and ships only the branding page independently while that question is unresolved. That is a deliberate scope decision, not an oversight: the pages above assume answers (a support-tier taxonomy, a claims policy) that no canonical file in this repository states today.

If your question is closer to “are we ready to turn a pilot into a program” — support tiers, what an enterprise IT function will ask for — there is no production-readiness matrix here yet. Usage and cost for a period is the one part of that question the repository does answer today: GET /api/v1/admin/usage (admin-only) aggregates inference_routing_log by user, provider, model, tier or day over a date_from/date_to range and returns request counts, token sums and an estimated cost per group plus deployment-wide totals — see docs/api/backend-openapi.yaml. Refused calls are excluded from the aggregation by default. Beyond that, the nearest honest answers live scattered across the trust centre (what is measured and evidenced), Operate (what is actually running and how you back it up), and Published gaps (what is explicitly not built). None of those is a substitute for a readiness matrix; treat this paragraph as a pointer, not an answer.

Forking as a multi-tenant SaaS

Section titled “Forking as a multi-tenant SaaS”

LQ.AI’s own code is Apache 2.0 (see branding and licensing), and Apache 2.0 does not prohibit running a fork as a hosted, multi-tenant service. That makes a fork-based SaaS offering legal — for LQ.AI’s own code. The web client is not Apache 2.0: web/LICENSE clause 4 binds any deployment or distribution, and a multi-tenant service is the likeliest way to pass its fifty-end-user / rolling-30-day threshold. Above the line you keep OpenWebUI’s branding alongside yours, or you hold written permission or an enterprise license from OpenWebUI. Read branding and licensing obligations before you price the offering. A fork-based SaaS offering is not, however, something this project builds or supports: the Inference Gateway’s key model assumes one operator holding provider keys for their own deployment, and billing individual end users to their own provider accounts is, in the architecture’s own words, “a different architecture” (docs/security/encrypted-keys.md, citing ADR 0011). Nothing in the codebase, the PRD, or this site documents how to build that architecture. If you are evaluating it, read that as a real gap, not a hidden feature.

Next

Section titled “Next”