Skip to content

Contextual Intelligence in Ultimus

Context is the foundation of agentic action. It is what a system knows before it executes: your code, your policies, and the history of its previous runs. Ultimus aggregates, verifies, and serves this context to both autonomous agents and the humans managing them.

02

Four-Dimensional Evaluation

Ultimus checks every agent decision against four core criteria:

Safety
Cryptographically verifying that the action is permitted by your policies.
Reliability
Confirming the output is accurate and passes your automated tests.
Efficiency
Tracking the exact cost in tokens, dollars, and estimated carbon output.
Utility
Measuring whether the action successfully delivered the intended business outcome.

03

Unified Telemetry

Agents and human operators rely on the same verifiable telemetry.

For agents

{
  "scope": "31 sessions, Ultimus on an open-weight model",
  "output_tokens": 593257,
  "cost_usd": 0.83,
  "gco2e_per_1k_output": 0.35,
  "carbon_basis": "estimated"
}

Agents receive structured JSON payloads (e.g., tracking token counts and precise costs) to dynamically route tasks to cheaper or more efficient models.

For people

593,257
Output tokens
$0.83
Total cost
0.35 g
CO₂e per 1k output tokens, estimated

Human Operators receive clean audit dashboards showing aggregate tokens, total cost (e.g., $0.83 for 593k tokens), and estimated carbon footprint, enabling data-driven budget and workflow adjustments.

The figures are real, from the Ultimus cost benchmark; the format is illustrative. Cost is measured; carbon is an estimate. See how they were produced

Start here

Add a module to what you run, or let us run it for you.

Zephyr, the signing and verification authority, is available on the CDAO Tradewinds Solutions Marketplace.