MetaGate

AI permission checks for DataHub assets.

Decision context

Before the agent acts

Choose an asset and action. MetaGate checks the evidence first.

InputDataHub asset + requested AI action.
DecisionAllowed or blocked with evidence.
If blockedCopy the repair and re-check.
Ready to check a DataHub asset
Connection Source: local DataHub Review: MetaGate API Not connected Waiting for DataHub. Scope: checking configured assets... Runtime: checking process identity... Ready.
Decision

Current check

Waiting for a current DataHub evaluation.

Waiting
The decision will appear after an asset is evaluated.
—datasets in the current run
—blocked autonomous actions
—allowed autonomous actions
30policy tests passed
History

Recent checks

Saved evaluations from this catalog.

0 assets
Hive

DataHub dataset

Repair plan

What failed, who owns it, and what to change.

blocked
Risk

Owner

Data owner

Repair

Rerun

Expected unlock

Decision
Readiness
Confidence
Why this matters

Next human action

Blocking evidence

    Evidence used

      Evidence coverage

      Fix the block

      Repair plan

      Waiting for an evaluation.

      Waiting

      The plan appears after a blocked evaluation.

      Technical receipts

      Proof & audit

      Agent permissions, integration checks, score math, and the decision trace.

      Enforcement receipt

      Agent permissions

      unavailable
      Permitted actionWaiting for evidence
      Human approvalUnknown
      EvidenceNot checked
      Decision IDNot recorded

      The contract appears after the selected asset is evaluated.

      Integration checks

      waiting

      Run a connected evaluation to compare the agent entry points.

      AgentWaiting
      SkillWaiting
      ToolWaiting
      ServiceWaiting

      The agent, skill, tool, and service must be registered and linked before a governed action can run.

      Skill pathWaiting
      MetaGate MCPWaiting
      AgreementWaiting

      The official DataHub MCP remains a separate optional server; this panel does not pretend it ran unless it is configured.

      Official DataHub MCP trace 
      No MCP call recorded.

      Repair loop proof

      Waiting

      Repair, indexing check, and re-check appear here.

        Adversarial gate

        Waiting

        Generated gate cases appear here.

        Show representative cases 
        
              
        
            
        
          
        
          
        
            
        Score basis
        
            
        This score will show the DataHub evidence used to calculate it.
        
            
        
              Evidence math
              
        
            
        
          
        
          
        
            
        Assessment profile
        
            
        The profile determines which checks matter for this kind of dataset.
        
            
        
            
        
            
        
            
        
              Dataset rubric
              
        
            
        
          
        

          
        
            
        
              
        
                Decision rule
                
        
              
        
            
        
            
        
              
        Readable decision
        
              
        
                
        
                
          
              
          
            
          
          
          

          
          
            How MetaGate reaches this decision
            
          
              
          1Read DataHub
          MetaGate gathers ownership, lineage, definitions, quality, freshness, usage, and assertion evidence.
          
              
          2Apply policy
          The requested action is checked against explicit metadata-backed terms.
          
              
          3Return decision
          The result is available here for humans and as JSON for agents or CI.
          
            
          
          
          
          
          

          
          
            
          Hackathon DataHub resources
          
            
          Load one of these resources into DataHub. MetaGate can then find and review the real dataset URNs.
          
            
          
              
          
                
          DataHub Docs
          
                
          Reference material for setup, agents, and platform behavior.
          
                Open DataHub Docs
              
          
              
          
                
          DataHub Quickstart
          
                
          Run the local DataHub platform used by the live proof.
          
                Open Quickstart
              
          
              
          
                
          DataHub MCP Server
          
                
          Let an agent read catalog context through a governed interface.
          
                Open MCP repository
              
          
              
          
                
          Agent Context Kit
          
                
          Give agents structured metadata context before they act.
          
                Open Context Kit docs
              
          
              
          
                
          DataHub Skills
          
                
          Reusable catalog skills for metadata-aware agent workflows.
          
                Open Skills docs
                Repository
              
          
              
          
                
          DataHub Core
          
                
          The open-source catalog platform that stores the graph MetaGate evaluates.
          
                Open Core repository
              
          
              
          
                
          Showcase Ecommerce
          
                
          Rich cross-platform graph with lineage, governance, glossary, and domains.
          
                datahub datapack load showcase-ecommerce --force
                Load instructions
              
          
              
          
                
          Bootstrap
          
                
          Small starter graph for a quick end-to-end smoke test.
          
                datahub datapack load bootstrap --force
                Load instructions
              
          
              
          
                
          NYC Taxi
          
                
          ~85 MB databases. Raw -> staging -> mart, with a planted freshness issue.
          
                Open dataset recipe
              
          
              
          
                
          Healthcare
          
                
          ~2 MB database. Raw -> staging -> billing + demographics, with planted quality issues.
          
                Open dataset recipe
              
          
              
          
                
          Fiction Retail
          
                
          ~95 MB database. Ten flat tables for orders, fulfillment, and returns.
          
                Open dataset recipe
              
          
              
          
                
          Analytics Agent
          
                
          DataHub’s analytics-agent path for metadata-aware analysis.
          
                Open Analytics Agent docs
              
          
              
          
                
          DataHub community
          
                
          Ask questions, get feedback, and learn from maintainers.
          
                Join hackathon Slack
                Town Halls
              
          
            
          
            
          
              
          Why these resources matter
          
              
          
                DataHub use Live GraphQL context, MCP/agent paths, and catalog discovery.
                Technical execution CLI, API, review surface, browser panel, and tests over metadata.
                Originality A capability-specific gate before an AI action, not another catalog view.
                Real-world value Freshness, quality, lineage, ownership, and incident evidence become explainable decisions.
              
          
            
          
            
          
              
          Assets currently in this DataHub
          
              
          Only assets actually found in DataHub are scored. A scenario stays pending until its metadata is loaded and indexed.
          
              
              
          Connect a live local review server to search the catalog.
          
            
          
          
          

          
          
            Proof & audit
            
          
            
          
              
          Capability Matrix
          
              
          
                
                  
          AI actionDecisionWhy
          
                          
                
              
          
            
          

            
          
              
          Capability Diff
          
              
          
                
          
                  Before repair
                  
            
                
            
                
            ->
            
                
            
                  After repair
                  
              
                
              
              
              
            
              

            
              
              
              Decision History
              
              
              Checks made during this review session, plus the next step.
              
              
              
            
              

            
              
              
              Human review
              
              
              Record whether this decision looks right. Local review saves to the review server; the hosted demo keeps a browser copy.
              
              
              
                
                
              
              
              
              
              No review recorded yet.
              
            
              

            
              
              
              Security & RBAC
              
              
              
                
              Read-only defaultMetaGate evaluates DataHub metadata without sending mutations unless write-back is explicitly configured.
              
                
              Server-side tokenDataHub tokens stay in the private MetaGate API, not in the public page or browser panel.
              
                
              Separated rolesRequester asks for an AI action, MetaGate evaluates policy, metadata owner repairs evidence, approver enables write-back.
              
                
              Mutation gateLive write-back requires deployment-approved GraphQL mutation documents.