An agent is only as good as what it knows. Behind every agent runs a stack of services that build a picture of your business, watch your market, and learn from what works.
These services run on their own — you do not invoke them. They are the difference between an agent that writes generic copy and one that writes like it has worked at your company for a year.
The business brain
Four layers combine into the context injected at the start of every agent task. Together they answer: who is this business, and how does it sound?
Archetype
Your business is matched to a universal business archetype — a proven playbook for how a company of that shape wins work. It gives every agent a strategic starting point instead of a blank page.
Business profile
A living profile — your competitors, your strongest hooks, the pain points you solve. It is not filled in once; it is kept current as the system learns more.
Voice engine
From real samples of your writing, the voice engine learns how you actually sound — so a drafted email reads like you wrote it, not like a template.
Knowledge base
Your own documents — PDFs, proposals, transcripts, URLs — ingested, chunked and embedded, so agents can pull exact facts from your material instead of approximating.
Why setup is gated
Until the core of this brain is in place — philosophy, brand voice, audience — the agents will hold back on outward-facing work. Generic output sent to real prospects costs more than the wait. See Business setup for what to fill in.
Watching your market
The signal monitor watches the sources you point it at — competitor sites, pricing pages, key URLs — on a schedule. When something changes, it does not just log it; it turns the change into a concrete opportunity an agent can act on.
Signal monitor
A competitor drops their price, ships a new service, or rewrites a landing page — the monitor catches it and surfaces it as an opportunity, so you respond in days, not when you happen to notice.
The learning loop
Nothing the agents do is fire-and-forget. The performance learning service reads the results of finished work — opens, replies, conversions, what was sent and when — and pulls out the patterns.
Performance learning
It analyses campaigns on a schedule, extracts what correlated with results, and feeds those lessons back into the business profile — and pings the Project Manager for a proper retro. Each round of work makes the next round sharper.
The engine room: model routing
Every agent task is, underneath, one or more calls to a large language model. Rigyx does not send all of them to the same place.
Task classes
Each call is tagged by what it needs — a quick classification, a quality draft, heavy strategic reasoning, or vision for browser control — and routed to the cheapest model that can do it well.
Tiers
Your plan tier sets the model line-up — a budget tier leans on fast, cheap models; a premium tier reaches for the top ones.
Bring your own keys
Supply your own provider API keys and the agents run on them. Keys are encrypted at rest; you are never locked to ours.
Spend caps
A monthly cap per business — a soft warning as you approach it, a hard stop at the limit. No surprise bills.
Where to go next
Architecture shows how these services, the agents and the model layer fit together as one running system.