Riviants

The problem

Small businesses get sold three things separately: a website builder, an SEO tool, and an ads platform. Each has its own login, its own jargon, and its own contradictions. The site builder doesn't know what the SEO tool is measuring; the SEO tool doesn't know what the ads platform is bidding on; the ads platform doesn't know what the site is converting. Owners patch the gaps with their attention — which they don't have — and the integration tax eats the result.

Riviants exists because the three problems are actually one: you need to ship a site that ranks and then keep paid acquisition aligned with what the site is doing. Treat that as one workflow, not three products.

The approach

Riviants is built as a coordinated agent stack rather than a feature pile. A page-generation agent ships layouts conditioned on the business brief; an SEO agent scores every page against the keyword graph the business actually cares about and rewrites where the gap is closeable; a campaign agent reads conversion data from the live site and reweights spend across channels without a human re-checking dashboards.

The coordination layer is what makes it work. Each agent has typed inputs, typed outputs, and a memory of prior decisions — the same durability pattern Sophonix uses in client engagements. When the SEO agent decides to retarget a phrase, the page-generation agent sees the decision; when the campaign agent shifts spend, the site copy follows.

The build

Riviants is a multi-tenant SaaS in production. Pages are generated server-side and cached at the edge. The agent stack runs as a long-lived orchestration service with task queues that survive process restarts. Conversion events from each tenant's site loop back into the campaign agent within minutes, not nightly.

The user-facing surface stays deliberately simple: a single dashboard that shows what the agents decided and why, with override controls where a human wants the wheel. Owners don't learn three tools; they learn one.

What it does in production

Riviants is live, in-market, and powers customer sites today. Architectural decisions about how to coordinate the agents directly inform how Sophonix scopes client engagements — the same pattern of “small focused agents + typed coordination + persistent memory” is the shape of every Sophonix delivery.

Tech stack

• Application: TypeScript + Next.js, edge-cached static + server-rendered hybrid
• Agent orchestration: persistent task queue with typed agent boundaries
• Data layer: relational store for tenant config, vector store for keyword and content recall, graph store for decision lineage
• LLMs: model-routed depending on task (capability vs cost tradeoff per agent)
• Hosting: bare-metal EU infrastructure, no third-party tracker dependencies on the customer page surface unless the customer opts in