AI Multi-Provider Consensus Gate in Kestra

id: ai-multi-provider-consensus-gate
namespace: company.governance

description: |
  Evaluate high-stakes operational requests across multiple independent frontier AI
  providers in parallel, adjudicate verdict agreement and risk scores using strict JSON
  Schema, and enforce automated approval gates with Slack escalation on dissent.

triggers:
  - id: scheduled_compliance_sweep
    type: io.kestra.plugin.core.trigger.Schedule
    description: Periodic compliance evaluation sweep. Shipped disabled by default.
    cron: "0 8 * * 1-5"
    disabled: true

  - id: governance_webhook
    type: io.kestra.plugin.core.trigger.Webhook
    description: Webhook triggered by change management systems (Jira, ServiceNow, GitOps).
    key: "{{ secret('GOVERNANCE_WEBHOOK_KEY') }}"

inputs:
  - id: request_id
    type: STRING
    defaults: CHG-89214
    description: Unique change management or operational request ticket identifier.

  - id: decision_context
    type: STRING
    defaults: "Emergency off-hours schema migration: Adding a non-null column with
      default lock on production customer_payments table during low-traffic
      window."
    description: Detailed context of the proposed operational change or transaction
      under review.

  - id: min_confidence_threshold
    type: FLOAT
    defaults: 0.80
    description: Minimum average confidence required across both providers to grant
      automated approval.

  - id: primary_model
    type: STRING
    defaults: gpt-4o-mini
    description: Primary frontier LLM provider model identifier.

  - id: secondary_model
    type: STRING
    defaults: gemini-2.5-flash
    description: Secondary independent LLM provider model identifier.

tasks:
  - id: evaluate_providers_in_parallel
    type: io.kestra.plugin.core.flow.Parallel
    description: Dispatches the decision payload concurrently to two distinct
      frontier AI providers.
    tasks:
      - id: evaluate_with_openai
        type: io.kestra.plugin.ai.completion.ChatCompletion
        description: Evaluates proposed operational change using OpenAI with strict JSON
          Schema output.
        provider:
          type: io.kestra.plugin.ai.provider.OpenAI
          apiKey: "{{ secret('OPENAI_API_KEY') }}"
          modelName: "{{ inputs.primary_model }}"
        messages:
          - type: SYSTEM
            content: |
              You are an independent Chief AI Governance and Reliability Officer.
              Evaluate the proposed operational change for safety, compliance risk, and potential service disruption.
              Assign a definitive verdict:
              - APPROVE: Safe to execute automatically.
              - REJECT: Unsafe or violates architectural standards.
              - MANUAL_REVIEW: High risk, requires human on-call sign-off.

              Return valid JSON strictly matching the provided schema.
          - type: USER
            content: |
              Request ID: {{ inputs.request_id }}
              Context:
              {{ inputs.decision_context }}
        configuration:
          responseFormat:
            type: JSON
            jsonSchema:
              type: object
              properties:
                verdict:
                  type: string
                  enum:
                    - APPROVE
                    - REJECT
                    - MANUAL_REVIEW
                confidence:
                  type: number
                risk_score:
                  type: number
                compliance_concerns:
                  type: array
                  items:
                    type: string
                justification:
                  type: string
              required:
                - verdict
                - confidence
                - risk_score
                - compliance_concerns
                - justification

      - id: evaluate_with_gemini
        type: io.kestra.plugin.ai.completion.ChatCompletion
        description: Evaluates proposed operational change using Google Gemini with
          identical JSON Schema output.
        provider:
          type: io.kestra.plugin.ai.provider.GoogleGemini
          apiKey: "{{ secret('GEMINI_API_KEY') }}"
          modelName: "{{ inputs.secondary_model }}"
        messages:
          - type: SYSTEM
            content: |
              You are an independent Chief AI Governance and Reliability Officer.
              Evaluate the proposed operational change for safety, compliance risk, and potential service disruption.
              Assign a definitive verdict:
              - APPROVE: Safe to execute automatically.
              - REJECT: Unsafe or violates architectural standards.
              - MANUAL_REVIEW: High risk, requires human on-call sign-off.

              Return valid JSON strictly matching the provided schema.
          - type: USER
            content: |
              Request ID: {{ inputs.request_id }}
              Context:
              {{ inputs.decision_context }}
        configuration:
          responseFormat:
            type: JSON
            jsonSchema:
              type: object
              properties:
                verdict:
                  type: string
                  enum:
                    - APPROVE
                    - REJECT
                    - MANUAL_REVIEW
                confidence:
                  type: number
                risk_score:
                  type: number
                compliance_concerns:
                  type: array
                  items:
                    type: string
                justification:
                  type: string
              required:
                - verdict
                - confidence
                - risk_score
                - compliance_concerns
                - justification

  - id: adjudicate_consensus
    type: io.kestra.plugin.core.output.OutputValues
    description: Adjudicates cross-model agreement, calculates consensus matrix, and
      determines automated approval eligibility.
    values:
      openai_verdict: "{{ outputs.evaluate_with_openai.jsonOutput.verdict }}"
      gemini_verdict: "{{ outputs.evaluate_with_gemini.jsonOutput.verdict }}"
      consensus_reached: "{{ outputs.evaluate_with_openai.jsonOutput.verdict ==
        outputs.evaluate_with_gemini.jsonOutput.verdict }}"
      average_confidence: "{{ (outputs.evaluate_with_openai.jsonOutput.confidence +
        outputs.evaluate_with_gemini.jsonOutput.confidence) / 2 }}"
      average_risk: "{{ (outputs.evaluate_with_openai.jsonOutput.risk_score +
        outputs.evaluate_with_gemini.jsonOutput.risk_score) / 2 }}"
      auto_approval_granted: "{{ outputs.evaluate_with_openai.jsonOutput.verdict ==
        'APPROVE' and outputs.evaluate_with_gemini.jsonOutput.verdict ==
        'APPROVE' and ((outputs.evaluate_with_openai.jsonOutput.confidence +
        outputs.evaluate_with_gemini.jsonOutput.confidence) / 2) >=
        inputs.min_confidence_threshold }}"

  - id: evaluate_approval_gate
    type: io.kestra.plugin.core.flow.If
    description: Grants automated pass-through on unanimous consensus approval or
      diverts to Slack human escalation.
    condition: "{{ outputs.adjudicate_consensus.auto_approval_granted == true }}"
    then:
      - id: export_approval_audit_trail
        type: io.kestra.plugin.core.storage.Write
        description: Persists certified multi-model consensus audit artifact in Kestra
          storage.
        fileName: "consensus-approval-{{ inputs.request_id }}.md"
        content: |
          # Multi-Model AI Consensus Approval Audit Trail

          - **Request ID:** `{{ inputs.request_id }}`
          - **Certified At:** `{{ now() }}`
          - **Consensus Verdict:** `APPROVED`
          - **Average Confidence:** `{{ outputs.adjudicate_consensus.average_confidence }}`
          - **Average Risk Score:** `{{ outputs.adjudicate_consensus.average_risk }}`

          ## Provider Evaluations
          - **OpenAI ({{ inputs.primary_model }}):** `{{ outputs.adjudicate_consensus.openai_verdict }}` (Confidence: {{ outputs.evaluate_with_openai.jsonOutput.confidence }})
            - *Justification:* {{ outputs.evaluate_with_openai.jsonOutput.justification }}
          - **Gemini ({{ inputs.secondary_model }}):** `{{ outputs.adjudicate_consensus.gemini_verdict }}` (Confidence: {{ outputs.evaluate_with_gemini.jsonOutput.confidence }})
            - *Justification:* {{ outputs.evaluate_with_gemini.jsonOutput.justification }}

      - id: log_consensus_approval
        type: io.kestra.plugin.core.log.Log
        description: Logs successful multi-model consensus verification.
        message: "Request {{ inputs.request_id }} approved by unanimous AI consensus
          (avg confidence: {{ outputs.adjudicate_consensus.average_confidence
          }}). Downstream execution authorized."

    else:
      - id: export_dissent_report
        type: io.kestra.plugin.core.storage.Write
        description: Persists side-by-side model divergence report for human review.
        fileName: "consensus-dissent-report-{{ inputs.request_id }}.md"
        content: |
          # Multi-Model AI Governance Escalation Report

          - **Request ID:** `{{ inputs.request_id }}`
          - **Escalated At:** `{{ now() }}`
          - **Consensus Reached:** `{{ outputs.adjudicate_consensus.consensus_reached }}`
          - **OpenAI Verdict:** `{{ outputs.adjudicate_consensus.openai_verdict }}`
          - **Gemini Verdict:** `{{ outputs.adjudicate_consensus.gemini_verdict }}`

          ## Side-by-Side Model Reasoning
          ### OpenAI ({{ inputs.primary_model }})
          - **Verdict:** `{{ outputs.evaluate_with_openai.jsonOutput.verdict }}`
          - **Confidence:** {{ outputs.evaluate_with_openai.jsonOutput.confidence }} | **Risk Score:** {{ outputs.evaluate_with_openai.jsonOutput.risk_score }}
          - **Concerns:** {{ outputs.evaluate_with_openai.jsonOutput.compliance_concerns | json }}
          - **Reasoning:** {{ outputs.evaluate_with_openai.jsonOutput.justification }}

          ### Gemini ({{ inputs.secondary_model }})
          - **Verdict:** `{{ outputs.evaluate_with_gemini.jsonOutput.verdict }}`
          - **Confidence:** {{ outputs.evaluate_with_gemini.jsonOutput.confidence }} | **Risk Score:** {{ outputs.evaluate_with_gemini.jsonOutput.risk_score }}
          - **Concerns:** {{ outputs.evaluate_with_gemini.jsonOutput.compliance_concerns | json }}
          - **Reasoning:** {{ outputs.evaluate_with_gemini.jsonOutput.justification }}

      - id: notify_compliance_slack_escalation
        type: io.kestra.plugin.slack.notifications.SlackIncomingWebhook
        description: Alerts compliance and platform on-call channel with side-by-side
          model reasoning for human sign-off.
        url: "{{ secret('SLACK_WEBHOOK_URL') }}"
        payload: |
          {
            "text": "⚖️ *AI GOVERNANCE ESCALATION: Request {{ inputs.request_id }}*\n*Decision Divergence Detected!*\n*OpenAI:* `{{ outputs.adjudicate_consensus.openai_verdict }}` (Confidence: {{ outputs.evaluate_with_openai.jsonOutput.confidence }})\n*Gemini:* `{{ outputs.adjudicate_consensus.gemini_verdict }}` (Confidence: {{ outputs.evaluate_with_gemini.jsonOutput.confidence }})\n*Auto-Approval:* `BLOCKED`\n*Dissent Report:* `{{ outputs.export_dissent_report.uri }}`\n*Action:* Human on-call review required before change execution."
          }

errors:
  - id: alert_governance_failure
    type: io.kestra.plugin.slack.notifications.SlackIncomingWebhook
    description: Alerts Slack on-call if any AI provider API or evaluation task errors.
    url: "{{ secret('SLACK_WEBHOOK_URL') }}"
    payload: |
      {
        "text": "⚠️ *AI Governance Flow Failure:* Execution {{ execution.id }} on flow {{ flow.id }} encountered an error communicating with providers."
      }

Diagram unavailable

We could not build the topology for this blueprint. The flow itself is valid, use the YAML on the left to run it.

When automating operational approvals (such as production database schema migrations, IAM permission escalations, or large transaction dispute refunds), relying on a single LLM is a major risk. A single model can hallucinate subtle syntax hazards (such as default-lock DDL on high-traffic tables), miss architectural guidelines, or suffer provider-specific reasoning quirks.

This blueprint orchestrates multi-provider AI consensus in Kestra: it concurrently dispatches candidate change requests to two independent frontier AI providers (OpenAI and Google Gemini) in parallel with identical JSON schemas. If both providers reach unanimous consensus with high confidence, the change passes automatically. If the models split or disagree, the workflow halts execution, formats an executive side-by-side comparison report, and alerts on-call engineers in Slack for human sign-off.

How it works

evaluate_providers_in_parallel (io.kestra.plugin.core.flow.Parallel) queries OpenAI and Google Gemini simultaneously with identical JSON Schema requirements.
evaluate_with_openai (io.kestra.plugin.ai.completion.ChatCompletion) generates a structured verdict (APPROVE, REJECT, MANUAL_REVIEW), confidence score, and justification using OpenAI.
evaluate_with_gemini (io.kestra.plugin.ai.completion.ChatCompletion) independently evaluates the same request using Google Gemini.
adjudicate_consensus (io.kestra.plugin.core.output.OutputValues) computes cross-provider agreement, calculates average confidence, and determines automated approval status.
evaluate_approval_gate (io.kestra.plugin.core.flow.If) branches deterministically:
- Consensus Approval: Persists an immutable approval audit trail artifact (consensus-approval.md) in Kestra storage and authorizes downstream execution.
- Dissent / Low Confidence: Halts automated execution, generates a side-by-side comparison report (consensus-dissent-report.md), and delivers an actionable escalation card to Slack for human sign-off.
alert_governance_failure (errors block) notifies Slack if provider communication fails.

What you get

Zero single-model dependency for mission-critical enterprise workflows.
Multi-cloud parallel orchestration executing in sub-seconds.
Publication-ready audit artifacts stored in Kestra internal storage for SOC2/HIPAA compliance.
Interactive Slack escalation cards displaying side-by-side reasoning from multiple frontier models.

Who it's for

Platform Engineers and DevSecOps teams automating infrastructure change requests.
FinTech and Banking compliance teams requiring multi-party verification for automated transactions.
AI Safety and Governance teams enforcing European AI Act oversight mandates.

Why orchestrate this with Kestra

Coordinating multiple independent cloud AI providers concurrently requires robust parallel execution, structured output validation, state persistence, and human escalation channels. Kestra handles multi-cloud parallelism, retries, and execution lineage natively without custom glue code.

Prerequisites

OpenAI API key.
Google Gemini API key.
Slack incoming webhook URL for escalation notifications.

Secrets

OPENAI_API_KEY: API key for OpenAI provider evaluation.
GEMINI_API_KEY: API key for Google Gemini provider evaluation.
SLACK_WEBHOOK_URL: Slack incoming webhook endpoint for governance escalations.
GOVERNANCE_WEBHOOK_KEY: Authentication key for event-driven CI/CD triggers.

Quick start

Configure OPENAI_API_KEY, GEMINI_API_KEY, and SLACK_WEBHOOK_URL in your Kestra namespace.
Import this blueprint into your Kestra instance.
Click Execute with default parameters to see parallel multi-model adjudication in action.
Enable the schedule or trigger the webhook from Jira/ServiceNow change approval workflows.

How to extend

Add Anthropic Claude (io.kestra.plugin.ai.provider.Anthropic) as a third voting party for 2-out-of-3 supermajority quorum.
Chain downstream cloud provisioning tasks (Terraform, AWS, Ansible) upon consensus approval.
Link human Slack reaction buttons to resume execution dynamically.

AI Multi-Provider Consensus Gate and Governance Adjudicator

How it works

What you get

Who it's for

Why orchestrate this with Kestra

Prerequisites

Secrets

Quick start

How to extend

Links

Tasks

Orchestrate with Kestra

Share this Blueprint

Create a Blueprint

New to Kestra?

AI Multi-Provider Consensus Gate and Governance Adjudicator

How it works

What you get

Who it's for

Why orchestrate this with Kestra

Prerequisites

Secrets

Quick start

How to extend

Links

Tasks

Orchestrate with Kestra

Share this Blueprint

Create a Blueprint

More AI Blueprints

Summarize Text with an AI Agent, Generate Multilingual Summaries in Short, Medium, or Long Format

AI Customer Call Compliance and Regulatory Disclosure Auditor

Trigger the Right Automation with an AI Agent Based on User Intent

Automate Customer Feedback Triage, Sentiment Analysis, and Escalation with Google Gemini and Slack

Generate AI Blog Posts in Your Brand Voice and Publish to WordPress as Draft

AI Annual Report Finder and Data Strategy Summary

More Core Blueprints

AI Customer Call Compliance and Regulatory Disclosure Auditor

AI Data Quality Anomaly Advisor and SQL Remediation Generator

Advanced Scheduling Automation, Run Tasks Only on Business Days by Country

Convert Natural Language Prompts to Structured Todoist Tasks with Deepseek AI

Invoice Math Discrepancy Gate with In-Memory DuckDB Reconciliation

AI Contract Clause Risk Analyzer and Legal Sign-Off Gate

New to Kestra?