A person stays in charge
How the repeatable work stays useful
Orchestration
One focused job, a parent waiting on a result, and no swarm fighting over the GPU.
Routing
The right model for the task — cheaper, safer, and still under policy.
RAG
Agents quote your brand kit and facts pack instead of inventing a company.
LLM Ops
Nothing goes live until a human unlocks it. Evaluation is the product.
Agents do the repeatable work. You keep the goals, the brand, and the spend. This page is how that stays true: who does the job, which model may answer, what they may cite, and the person who decides it is live.
The names are orchestration, routing, RAG (they look up house truth before they write), and a human in the loop. Those are the method. They are not the offer.
How we work · Platform · Blog · Contact
Words we use on purpose
- Agentic-first — humans own goals, brand, and spend; agents execute scoped work under policy
- Orchestration — roles, handoffs, recovery, one job at a time
- Routing — which model and which machine answers, given privacy, cost, and skill
- RAG — retrieval so agents cite the brand kit, not a plausible fiction
- LLM Ops — observe, evaluate, gate, and promote model work with humans in the loop
- HITL / dry-run — preview until a human unlocks; nothing auto-publishes
- Sole-leaf execution — one focused task, so the GPU and the team are not thrashed
Orchestration
What breaks: Agent armies redo work, open finished tickets, and burn the only GPU.
What you get: Specialist roles under a human Board, a parent blocked on the active child, and recovery that does not reopen done work.
Read: Agentic-first companies · Local orchestration
Routing
What breaks: One model for every task leaks data or wastes spend.
What you get: Task class maps to local, remote, or human. Routes fail closed when they violate policy.
Read: LLM routing, data centers & RAG
RAG
What breaks: Agents invent a company — hex codes, legal notices, founder bios.
What you get: Ranked retrieval over Board-gated facts, approved pages, and runbooks, plus a forbidden-claim filter at generate time.
Read: Brand kits as source of truth · RAG as corporate memory
LLM Ops
What breaks: Chat demos do not survive a real publish path.
What you get: Goldens, forbidden-claims, dry-run, and Board unlock. Promotion is a protocol. No agent real-money authority.
Read: ML evaluation loops · Evaluation protocol · Dry-run by default
How the value flows
A human sets the goal. Orchestration assigns one specialist. Routing picks the model. RAG supplies house truth. Ops gates hold the result until a person unlocks it.
Presence is this stack applied to corporate presence — the living lab, not the whole firm.
Next
- How we work · 2. Platform · 3. Blog · 4. Contact
*Preview and dry-run until Board unlock. No unsupervised publish-without-review flows.*