Advisory
A facilitated half-day for the group that controls the budget. We walk the seven-stage context supply chain against your estate, place you on the L0–L5 maturity scale with reasoning shown, and cost the options — including doing nothing. Bring the skeptic.
- One causal story of where context breaks, examined by the whole room
- Directional L0–L5 placement, reasoning written down
- The next engagement costed honestly — with what it would not fix
- One-page summary for stakeholders who were not in the room
Diagnostics
The entry diagnostic — approvable without a steering committee. Two weeks of read-only telemetry work producing cost per unit of work, a quantified waste analysis, and a remediation list with modeled recovery. Built so a finance reviewer can reproduce the arithmetic without calling us.
- Cost per unit of work, not cost per million tokens
- Five waste patterns hunted and quantified in your telemetry
- Recovery modeled as ranges with assumptions stated — never guaranteed
- Ends with a clear go or no-go for the Assessment, including “your spend is fine”
Diagnostics
Measurement across all seven stages: sources sampled for freshness, retrieval tested against ground truth, semantics checked for drift, and the learning loop asked when a failure last changed anything. Diagnose establishes what is true; Blueprint decides what to do — and you can take the first without the second.
- L0–L5 scorecard with evidence notes on every score
- Targets set per stage per use case — not L5 everywhere
- Implementation quarters costed before you fund them
- A written go or no-go — including “do not implement yet”
Delivery
The blueprint, built: structure, semantics, validation gates, retrieval with an evaluation harness, and the operational plumbing — in your environment, with your engineers embedded, so the capability stays when we leave. Scope and fee fixed at quarter start; runbooks are deliverables, not afterthoughts.
- Fee range tracks scope, priced by the blueprint before the quarter starts
- CoreModels as the authoring layer where the blueprint calls for it
- Acceptance criteria checked off item by item at handoff
- Built for your team to operate and extend without us
Operations
A named owner for the stage that quietly stops running. Monthly monitoring of source freshness and retrieval fitness, a living evaluation suite, incident attribution by layer, and a learning loop whose output is changes made — reported quarterly in ROKA terms a business reader can check.
- Retrieval quality re-tested on schedule against ground truth — regression becomes a dated event
- Every incident attributed: context, model, or application
- Failures become test cases; findings become recorded changes
- Quarterly ROKA report honest enough to say “flat” when it is flat
Software
Developed by ARAMAI · deployed and operated by Ariesnet.
Schema intelligence middleware. It is where the structure and meaning of your context get authored, governed and exported — in formats the rest of your stack already reads.
- Your teams keep their own vocabulary, and the systems still agree on what a customer is
- The conformance contract exports in formats your stack already reads — JSON Schema, JSON-LD, XSD, ShEx
- Your agents connect over MCP on paid plans, running on your own agent API key
- Start free; hitting a plan limit never blocks work you have already done
