Kanvaz AI Labs

Platform

Modular components.
Adopt one, or all.

Read-only middleware between your source systems and planning. Start with a single component; your existing stack stays.

01 — 7 items

GlossDB Hub

ComponentAvailable

Declared schema

The shared planning vocabulary — units, metrics, measurements — declared up front. Nothing is written against an undeclared aspect.

  • — glossQL DECLARE ASPECT
  • — Facts, queries and column measurements
  • — Versioned with every change
ComponentAvailable

Data contracts

Each metric carries its SQL and every assumption with a confidence score, so planning knows exactly what a number means.

  • — SQL + assumptions per metric
  • — Confidence per assumption
  • — Grounded on column statistics
ComponentAvailable

Mapping docket

Suggested mappings and unit conversions go to your experts once. Approved rulings are reused across loads and markets.

  • — Agent proposes, person rules
  • — Reusable across markets
  • — Accountable owner per ruling
ComponentAvailable

Load checks

Every new load is checked against approved rulings before it reaches planning, with deviations flagged by market.

  • — Per-load validation
  • — Deviation flags by market
  • — Data quality reports
ComponentAvailable

Bi-temporal replay

Every statement is a commit with author and data snapshot. Scrub back to see what planning was told at any point.

  • — Commit history per definition
  • — Valid time × recorded time
  • — Person / agent attribution
KitAvailable

Read-only connectors

Sits between source systems and your existing integration layer. Delivers in your current import format.

  • — Read-only by design
  • — No write-back into ERP or planning
  • — Flat-file or connector input
ComponentOpen source

glossQL

The open-source language GlossDB is built on.

  • — Open source
  • — SQL-adjacent syntax
  • — glossdb.com/docs

02 — 3 items

Kanvaz Causal Core

ComponentPilot
Aquila mark

Aquila

Causal discovery

Separates consumption signals from distortions caused by batching, promotions and shortage ordering.

  • — Structural causal models
  • — Multi-tier order amplification
  • — Driver attribution
ComponentPoC
Lynx mark

Lynx

Counterfactual evaluation

Compares interventions on planning policies and parameters before anything changes in production.

  • — do-calculus interventions
  • — Variance-reduction estimates
  • — Policy comparison
ComponentResearch
Steinbock mark

Steinbock

Safety envelope

No experiment runs unless it stays inside the combined safety envelope.

  • — Bounded experimentation
  • — Service-level guardrails
  • — Patent pending

03 — 3 items

Planner HUD & Workflows

ComponentPilot
Marmot mark

Marmot

Exception handover

Brings planning drift and operational exceptions into one focused daily view.

  • — Daily exception queue
  • — Explainable alerts
  • — Prioritised by impact
ComponentPilot
Chamois mark

Chamois

Governance levels

Agents may flag, propose, or act within agreed limits — set per decision type.

  • — Level 1–4 permissions
  • — Human approval for exceptions
  • — Full audit trail
ComponentPilot
Vulpes mark

Vulpes

Outcome loop

Measures what happened after each approved action and feeds learning into the next cycle.

  • — Decision-outcome tracking
  • — Parameter feedback
  • — Cycle-over-cycle learning

04 — 1 item

Bullwhip Sensor

KitPoC
Lepus mark

Lepus

Diagnostic kit

A 1–2 week read-only assessment of order amplification across your tiers.

  • — Flat-file exports only
  • — Variance ratio per tier
  • — Executive readout

See which business problems they solve.

Solutions · How we work

Book a briefing