KB Sandbox
Sign In

KB Sandbox

The governed enterprise AI Workbench: organize trusted knowledge, apply it to real business work, connect safely to live systems, and let regional builders extend it with evidence-backed tools and agents.

Explore ExamplesLearn How It WorksSign In

See what governed enterprise AI can do

Start with trusted knowledge. Grow into useful work, connected systems and new AI capabilities.

Follow the journey for an organization, a regional builder or a school AI laboratory. Every stage keeps evidence, permissions and human authority visible.

Begin with proof

Flagship demonstrations

Recognizable business problems first. Technical depth remains available inside each story.

Live demonstratedSecure HR knowledge

One document, two correctly different answers

Two active members ask the same question in the same Project. Only the authorized member receives the restricted evidence.

One restricted file · two users · zero restricted details leaked

Demonstration data: The employee names, policies and compensation figures in this walkthrough are synthetic and do not represent actual Sandz HR records or salary bands.

OrganizationsKnowledgeAccess controlEmber
View verified walkthrough
Live demonstratedEmber onboarding

From first question to a governed workspace

An ordinary employee describes Sandz in plain language. Ember creates the initial departmental structure and stops at the human-only access step.

Four turns · zero product jargon required · one human-only step
OrganizationsSchoolsOnboardingProjectsGovernance
Read the real conversation
Pilot demonstratedSales proposal copilot

Evidence to proposal—with people still in charge

Combine approved product evidence with customer requirements, identify what is missing and prepare a grounded proposal draft.

Cited capabilities · visible gaps · human commercial approval
OrganizationsBuildersSalesEvidenceApproval
Showcase in development

A progressive capability journey

Know → Apply → Connect → Build → Learn

The stages build on one another, but customers can start wherever their evidence, systems and people are ready.

1

Know

Build trusted organizational knowledge

Give people a useful AI interface to approved knowledge while preserving Project membership, document access and model-processing controls.

Live demonstratedSecure HR knowledge

One document, two correctly different answers

Two active members ask the same question in the same Project. Only the authorized member receives the restricted evidence.

One restricted file · two users · zero restricted details leaked

Demonstration data: The employee names, policies and compensation figures in this walkthrough are synthetic and do not represent actual Sandz HR records or salary bands.

OrganizationsKnowledgeAccess controlEmber
View verified walkthrough
Pilot demonstratedCustomer support copilot

Answer—or escalate instead of bluffing

Support staff receive cited product guidance and a clear escalation boundary when approved evidence does not settle the question.

Grounded answer · source citations · responsible escalation
OrganizationsSupportKnowledgeEscalation
Showcase in development
PrototypePolicy comparison

See what changed—and who needs to act

Compare two policy versions, identify material changes and trace each finding back to the relevant clauses.

Change summary · evidence map · human policy decision
OrganizationsSchoolsPolicyComparisonEvidence
Showcase in development
2

Apply

Turn evidence into useful work

Move beyond question-answering to proposals, comparisons and investigations without pretending that AI owns the final decision.

Pilot demonstratedSales proposal copilot

Evidence to proposal—with people still in charge

Combine approved product evidence with customer requirements, identify what is missing and prepare a grounded proposal draft.

Cited capabilities · visible gaps · human commercial approval
OrganizationsBuildersSalesEvidenceApproval
Showcase in development
PrototypeProcurement and RFPs

Compare vendors against one requirement set

Score candidate proposals against the buyer’s requirements rather than against each vendor’s marketing language.

Comparison matrix · unsupported claims · questions for review
OrganizationsSchoolsProcurementScoringHuman review
Showcase in development
PrototypeSemiconductor quality

Structure an investigation without inventing root cause

Organize incident evidence through an 8D investigation and expose the evidence still required before closure.

8D report · evidence map · quality-manager approval
OrganizationsBuildersSchoolsManufacturingInvestigationGovernance
Showcase in development
3

Connect

Bring in live systems and events

Knowledge explains what has been established. Connectors add current information; webhooks report what just changed.

ConceptCustomer ticket connector

Combine approved guidance with a live support case

Use governed product knowledge alongside live ticket status without copying an entire support system into the prompt.

Current case data · approved guidance · auditable handoff
OrganizationsBuildersAction ConnectorSupportREST API
Showcase in development
ConceptInvoice processing

Route invoice exceptions instead of automating judgment

Validate structured fields deterministically, compare supporting evidence and send discrepancies to the right person.

Validated fields · exception route · preserved audit trail
OrganizationsBuildersAccountingWebhookHuman gate
Showcase in development
ConceptOperations evidence

Collect logs without the copy-and-paste errors

Retrieve bounded diagnostic windows from monitoring platforms and return normalized evidence to an investigation Project.

Correlation IDs · bounded evidence · reproducible investigation
BuildersSchoolsAction ConnectorMetricsREST API
Showcase in development
4

Build

Create portable tools and agents

Use KB Sandbox to discover, specify and evaluate an integration; build and host it externally; expose only approved capabilities to Ember. Registration, certification and Project availability stay visibly distinct states — a registered capability is not yet connected, certified or available to Ember.

The builder lifecycle

  1. 01Business problem
  2. 02Discovery and Method
  3. 03Architecture and specification
  4. 04External implementation
  5. 05Registration and evaluation
  6. 06Human approval
  7. 07Available to Ember
PrototypeLegacy application to MCP

Turn a legacy application into an AI-ready service

Recover workflows and business rules, validate the API contract, then build and evaluate a narrowly scoped MCP server.

Evidence-backed specification · tested tools · versioned contract
BuildersSchoolsOpenAPIMCPModernization
See the discovery foundation
ConceptEveryday business

Order food through Ember—with confirmation

Builder-hosted outlet and delivery tools let Ember check options and prepare an order while the user confirms before spending.

Reusable MCP tools · explicit confirmation · portable service
OrganizationsBuildersSchoolsMCP ServerOrderingHuman confirmation
Showcase in development
ConceptExternal proposal agent

Add a specialist agent only when the workflow needs one

Coordinate research, approved tools and commercial gates through a versioned external agent when a bounded Ember tool loop is insufficient.

Declared graph · governed tools · evaluated human gates
BuildersExternal AgentMCPCertification
Showcase in development
5

Learn

Develop the next generation of AI builders

Turn genuine regional problems into evidence, specifications, deployed capabilities and portfolio-quality demonstrations.

School AI Builder Laboratory

  1. 01Learn the foundations
  2. 02Choose a real regional problem
  3. 03Create evidence and architecture
  4. 04Build with assisted coding
  5. 05Register and test the capability
  6. 06Deploy on approved infrastructure
  7. 07Demonstrate it through Ember
ConceptSchool AI Builder Laboratory

From student idea to deployed AI capability

Students learn the foundations, specify a real problem, build with assisted coding, test the result and demonstrate it through Ember.

Evidence · architecture · implementation · tests · deployment
SchoolsBuildersEducationHackathonDeployment
Showcase in development
Live demonstratedReproducible AI experiment

Compare models on the same evidence and method

Students and practitioners compare models or execution strategies without changing the benchmark halfway through.

Repeatable runs · comparable evidence · honest limitations
SchoolsBuildersEvaluationModelsReproducibility
View a published experiment

The operating principle

Discover with evidence. Specify the boundary. Build externally. Test the permissions. Approve the version. Let Ember use only what the Project needs.

Ready to see the complete published catalogue, or how KB Sandbox actually works?

Browse all ExamplesAbout KB Sandbox