Zenith is the Cortex inside RIOS — an autopilot built for a world where the wrong move costs more than the right one. It quotes, books, dispatches, documents, and settles within the exact boundaries of your policy, your contracts, and the law. Human-in-the-loop by default. Provable by construction.
Copilots suggest. Zenith operates — inside guardrails you can inspect, with logs you can replay, and stop conditions you can trust. It's the difference between "an LLM writing an email" and "a regulated system clearing a shipment."
Zenith can only act within typed, versioned policies — rate ceilings, carrier safety thresholds, HOS windows, insurance minimums, routing restrictions, customs lanes. Violating policy is not a bug; it's mechanically impossible.
Every decision is a typed event with inputs, model version, policy hash, and outputs. Replays are deterministic. Evidence bundles export in minutes for regulators, insurers, and counterparties.
Above configurable thresholds — dollar value, risk score, novelty — Zenith pauses for human sign-off. Below them, it moves.
Learns from outcomes inside guardrails. Every update is shadow-tested against replay sets before it ever touches production.
For any decision, Zenith will show the top alternatives it considered and why they lost. No black boxes.
Zenith runs a closed loop over every shipment. The inner loop is milliseconds. The outer loop is the lifecycle. Both are observable, both are stoppable, both leave evidence.
Typed events in, typed events out. Deterministic replay. Cryptographic evidence. No improv.
Zenith drives the six RIOS agents. Each one inherits Cortex guardrails — policy typing, confidence scoring, replay, human override — and each one can be enabled, tuned, or disabled independently per tenant.
Blended lane rates with confidence scores. Declines to quote below threshold rather than guessing.
Matches carriers against SLA, insurance, FMCSA safety score, and HOS window — no exceptions.
Weather delay, HOS break, capacity gap — Zenith revises the plan and logs the reason.
Merges ELD, GPS, and dock events into a single on-time story. Flags drift before it's a crisis.
BOL, POD, VUCEM, NOM, customs — generated from typed events, signed, and filed.
Tender-to-payout with evidence bundles. Disputes resolved in minutes, not months.
Autonomy in regulated industry is earned, not declared. These are the disciplines Zenith inherits from RIOS and why we trust it near real cargo, real money, and real people.
Given the same inputs, policy, and model version, Zenith produces the same decision. Full stop.
Every action is a typed event with a schema. If it doesn't typecheck, it doesn't run.
Rate ceilings, safety floors, insurance minimums — enforced at the Cortex layer, not the UI.
Every decision comes with inputs, alternatives, confidence, and the policy clauses that applied.
One-click disable — per agent, per lane, per customer. Failure modes are planned, not hoped for.
SOC 2 aligned. Regulator-ready export. On-chain attestation for the decisions that matter.
A Zenith booking is never a freeform prompt. It's a typed event — inputs, policy reference, model version, alternatives — replayable forever.
// Book a load under tenant policy v42 const decision = await rios.zenith.book({ load: "L-20260414-8821", policy: "tnt-acme@42", objectives: ["on_time", "cost"], constraints: { min_safety: 85, max_rate_pct_of_ceiling: 0.96, require_insurance: "CARGO_100K" }, human_above: { value_usd: 10000 } }); // → { action: "booked", carrier: "C-884", confidence: 0.93, // replay_id: "r_a1b2…", policy_hash: "0x7f…", alternatives: [...] }
// Replay any decision. Byte-for-byte. const replay = await rios.zenith.replay("r_a1b2…"); replay.inputs // → { load, lanes, ratesheets, safety, hos... } replay.policy // → { hash: "0x7f…", clauses: [...] } replay.model // → { name: "zenith-cortex", ver: "2.1.4" } replay.alternatives // → [{ carrier: "C-119", score: .88, rejected_by: "insurance" }, ...] replay.evidence // → { bundle_url, signature, timestamp }
Zenith ships inside EusoTrip. Enable it per lane, per agent, per tenant. Start narrow, prove it, widen the aperture.
