EcoIQ Platform Module

EcoIQ AI Agent Operations Console

Monitor agents, tasks, evidence, costs and human approvals.

EcoIQ AI Agent Operations Console gives teams a transparent control layer for AI agents. It shows agent status, queued tasks, running jobs, failed actions, evidence sources, model usage, estimated cost, confidence, safety gates and human approval requirements across EcoIQ's modernisation workflow.

Core purpose: Make EcoIQ's agent system observable, auditable, safe and enterprise-ready.

Open Agent Console View Running Tasks Review Failed Tasks Open Human Approval Queue View Evidence Trace Retry Failed Task Request Missing Evidence Export Agent Logs Send Approval to Teams Generate Morning Operations Briefing

Connected EcoIQ Modules

Amanah Autopilot

Runs the overnight supervisor agent this console monitors.

Omnimodal Evidence Panel

Supplies the photo/video evidence the Visual Evidence Agent analyses.

Product Analytics & KPI Engine

Consumes agent task metrics as part of platform-wide KPIs.

Command Centre

Surfaces which projects have active agent tasks running.

Governance & Expert Review Board

Receives tasks routed to the Human Approval Queue.

Data Room & Evidence Vault

Stores the evidence every agent output traces back to.

Asset Passport

Is created and updated by the Asset Passport Agent.

Impact MRV Layer

Is populated by the MRV Agent.

Institutional Finance Engine

Is populated by the Finance Modelling Agent.

Supplier & Funding Marketplace

Is populated by the Supplier / Funding Match Agent.

Executive Briefing & Board Pack Generator

Is populated by the Report Generator Agent.

API & Integration Layer

Exposes agent task and log data through the API.

Microsoft Ecosystem Core Stack

Provides the agent orchestration building blocks this console monitors.

Responsible AI tools

Support explainability, bias checking and safety review of agent outputs.

Teams approvals

Delivers human approval requests routed from the queue.

Power BI dashboards

Renders operations dashboards for agent activity and cost.

Microsoft Fabric

Stores agent logs and telemetry.

Agent Categories

Agent 1

Research Agent

Reads public documents, websites, reports and regulatory sources.

Tasks

Gather evidenceSummarise source dataIdentify missing documentsCreate source listFlag outdated information

Agent 2

Document Reader Agent

Reads PDFs, annual reports, energy bills, invoices, supplier quotes and technical files.

Tasks

Extract tablesDetect key factsClassify evidenceLink documents to assetsUpdate Data Room

Agent 3

Photo / Visual Evidence Agent

Analyses photos and videos from Mobile Inspection.

Tasks

Detect visible risksLabel asset componentsIdentify missing sensorsFlag safety concernsPrepare visual evidence notes

Visual findings are hypotheses and must show "Needs verification" unless confirmed by expert review or measurement.

Agent 4

Asset Passport Agent

Creates and updates asset records.

Tasks

Structure asset profileLink evidenceDetect missing baseline dataAssign asset typeRecommend playbook

Agent 5

Playbook Matching Agent

Matches assets to industrial modernisation playbooks.

Tasks

Identify best playbookRecommend quick winsSuggest deep upgradesCreate next action checklist

Agent 6

Finance Modelling Agent

Prepares financial models and investment logic.

Tasks

Estimate CAPEX/OPEXCalculate paybackPrepare finance memoIdentify funding gapFlag assumptions needing review

Agent 7

Supplier / Funding Match Agent

Matches projects to suppliers, funders and sponsors.

Tasks

Shortlist suppliersIdentify grant/CSR/funding routesPrepare RFQ packDraft outreachFlag due diligence gaps

Agent 8

MRV Agent

Tracks baseline, after-data and impact verification.

Tasks

Check baseline completenessCompare before/after evidenceFlag weak evidencePrepare MRV reportDistinguish estimated vs verified impact

Agent 9

Governance Agent

Prepares expert review packets and approval workflows.

Tasks

Identify required reviewersPrepare No Harm Gate checklistRoute to technical/financial/environmental/Maqasid reviewTrack approvals

Agent 10

Report Generator Agent

Creates investor memos, board packs, country briefs and public summaries.

Tasks

Generate executive summaryCite evidenceCreate decision memoPrepare appendicesFlag unsupported claims

Agent 11

Customer Success Agent

Monitors customers after sale.

Tasks

Detect at-risk accountsPrepare value reviewIdentify renewal riskSuggest upsellFlag missing MRV evidence

Agent 12

Sales CRM Agent

Supports sales and partner pipeline.

Tasks

Draft outreachPrepare follow-up remindersIdentify qualified leadsConnect proposals to productsFlag overclaiming risk

Agent 13

Analytics Agent

Monitors product and KPI performance.

Tasks

Detect funnel drop-offsPrepare KPI briefingIdentify high-performing countriesFlag weak module adoptionDetect MRV bottlenecks

Agent 14

Amanah Autopilot Supervisor

Runs overnight checks across the platform.

Tasks

Detect high-harm assetsFind missing evidencePrepare morning briefingIdentify finance-ready projectsFlag approvals neededPrepare human review queue

Task Statuses

QueuedRunningWaiting for EvidenceWaiting for Human ApprovalNeeds VerificationCompletedFailedBlockedEscalatedCancelled

Operations Dashboard Cards

Active agentsQueued tasksRunning tasksCompleted tasks todayFailed tasksBlocked tasksTasks waiting for evidenceTasks waiting for human approvalEstimated daily model costEvidence sources processedReports generatedMRV checks completedSupplier matches generatedFinance memos generatedNo Harm Gate alertsAverage task completion time

Task Table Fields

Task IDAgent nameProject / assetTask typeStatusPriorityStarted atLast updatedEvidence usedOutput generatedConfidence levelCost estimateModel usedError messageHuman approval requiredNext action

Agent Run Detail View

For every agent run show:

Prompt / instruction summaryInput dataEvidence sourcesDocuments readPhotos analysedModel usedTokens / estimated costOutput summaryConfidenceMissing dataSafety flagsNo Harm Gate flagsHuman approval statusFinal decisionAudit trail

Evidence Trace

Every agent output should show:

What evidence was used
Where it came from
When it was uploaded
Whether it is strong/medium/weak
Whether it is verified
Whether it is stale
Which claims depend on it
Which human reviewer approved it

Cost & Model Monitoring

Model usedTask typeToken usage estimateCost estimateCost by agentCost by customerCost by projectCost by product moduleHigh-cost tasksFailed-cost wasteCheaper model candidateSensitive-data model routing

Model routing examples

Sensitive client data → approved enterprise model / Azure OpenAI
Public data and reports → lower-cost model allowed
Complex finance memo → high-reasoning model
Image inspection → vision model
Long documents → long-context model
High-impact decision → human review required

Human Approval Queue

Show tasks waiting for:

Technical reviewFinance reviewEnvironmental reviewSafety reviewMaqasid/Mizan reviewIslamic finance reviewPublic summary approvalSupplier outreach approvalFunding memo approvalMRV verification approval

Failure and Debugging

Failed document parsing
Missing evidence
Model timeout
Invalid JSON output
Unsupported claim
Hallucination risk
Source mismatch
Route/API error
Permission error
Privacy/PII risk
Failed supplier match
Failed finance calculation
Blocked public reporting

Recovery actions

Retry taskRequest missing evidenceDowngrade claim to "Needs verification"Assign human reviewerSwitch modelReduce contextOpen Data RoomOpen evidence traceCreate support ticket

Example Agent Runs

Agent

Photo / Visual Evidence Agent

Needs Verification

Project: Boiler House #3

Input: 8 inspection photos

Findings

Uninsulated pipe visible
Soot marks visible
No smart meter visible
Corrosion risk possible

Next action: Request engineer review and collect 12 months fuel data.

Agent

Finance Modelling Agent

Waiting for Human Approval

Project: Factory Compressed Air Optimisation

Input: Energy bill, compressor photos, playbook match

Output

Draft payback estimate and investor memo outline.

Next action: Financial reviewer checks assumptions.

Agent

MRV Agent

Blocked

Project: Village Clean Heating Pilot

Input: Before photos, installation record, after comfort survey

Output

MRV report draft.

Blocker: After fuel data missing.

Next action: Request missing after-data before verified impact claim.

Agent

Amanah Autopilot Supervisor

Completed

Run: Overnight portfolio scan

Output

4 missing evidence alerts, 2 finance-ready projects, 1 public summary ready for review.

Amanah Autopilot Morning Operations Briefing

"Overnight, 42 tasks ran across 11 agents. 31 completed, 6 need evidence, 3 need human approval and 2 failed due to missing documents. Estimated model cost was £4.80. Top priority: Boiler House #3 finance memo needs technical review before supplier outreach."

Microsoft Integration

Azure AI Agent Framework / Semantic Kernel concepts for orchestration
Microsoft Fabric for agent logs and telemetry
Power BI for operations dashboards
Teams for human approval notifications
SharePoint / Data Room for evidence packs
Azure Monitor / Application Insights concept for tracing
Responsible AI Toolbox for explainability
Presidio for privacy / PII detection
PyRIT / red-team checks for agent safety

No Harm Gate for Agents

Before any agent output is used externally, check:

Is the evidence strong enough?
Is the output clearly labelled as draft/verified?
Is human approval required?
Does the agent overclaim savings, funding or impact?
Is personal or sensitive data protected?
Is public reporting approved?
Are religious/Maqasid/Mizan claims safe and not overstated?
Is model routing appropriate for sensitive data?
Is the task auditable?
Can the user trace every claim to evidence?

Safety and Governance

Agent outputs are decision-support drafts until reviewed and approved where required.
High-impact industrial, financial, environmental, safety, public reporting and Islamic finance decisions require human approval.
Visual inspection findings are hypotheses unless verified by measurement or expert review.
EcoIQ must not publish or act on unsupported claims.
Sensitive data must be routed, logged and protected according to permissions.
Maqasid/Mizan is ethical decision-support, not a fatwa.