EcoIQ Platform Module

EcoIQ AI Agent Council

Twelve operationally trained agents. Four next-stage agents. One governed intelligence system.

EcoIQ AI Agent Council presents how EcoIQ's specialised AI agents work together across industrial research, document reading, visual evidence, asset intelligence, playbook matching, finance, MRV, governance, reporting and overnight supervision. It shows which agents already have full operational training packs, which agents are next to be trained, how agents hand work to one another and where human approval is required.

Core purpose: Give investors, Microsoft ecosystem stakeholders, governments, industrial operators and partners one clear place to understand EcoIQ's AI agent architecture.

operational training packs

120

agent training files

master training index

next-stage agents

human approval gates

evidence-first workflows

Open AI Agent Council View Operational Agents View Agent Training Lab Open AI Agent Operations Console View Handoff Map Open Human Approval Flow View Amanah Autopilot Open Evidence Trace View Training Index Open Next-Stage Agents

Council Control Room — Live State

These are real, live-queried counts from this database, not illustrative numbers — every value reads 0 until python manage.py seed_council_demo_run has been run.

active council runs

open disagreements

decisions awaiting human review

low-confidence decisions (under 70%)

Council Runtime pages

Agent Training Lab Agent Reliability & Maturity Council Memory

Connected EcoIQ Modules

Agent Training & Evaluation Lab

Supplies the shared training method and evaluation approach every agent pack follows.

AI Agent Operations Console

Monitors live agent task runs, confidence and health across the Council.

Document Reader Agent Training Pack

The deep-dive training pack behind the Document Reader Agent seat.

MRV Agent Training Pack

The deep-dive training pack behind the MRV Agent seat.

Knowledge Graph & Relationship Map

Stores every agent handoff as an evidence-linked graph node.

Asset Passport

Receives the structured asset record Asset Passport Agent builds.

Industrial Playbook Library

Supplies the playbooks Industrial Playbook Matching Agent matches against.

Institutional Finance Engine

Receives Finance Modelling Agent draft CAPEX/OPEX models.

Impact MRV Layer

Receives MRV Agent baseline/after-data and verification status.

Governance & Expert Review Board

The human review layer Governance Agent routes packets to.

Executive Briefing & Board Pack Generator

Receives Report Generator Agent output for investor/board use.

Amanah Autopilot

The overnight supervision this Council relies on for portfolio-wide checks.

Data Room & Evidence Vault

Stores the evidence every agent handoff links back to.

Microsoft Ecosystem Core Stack

Supplies the Microsoft-ready building blocks the Council is designed to integrate with.

Operationally Trained Agents

12 agents currently have a full 10-file operational training pack in the repository (ai_agents/<agent_name>/), read live from the repository state — 120 agent training files in total.

Agent 1 Operational Training Pack Ready

Research Agent

Finds, compares and summarises trusted evidence.

Primary handoff

→ Document Reader Agent

Agent 2 Operational Training Pack Ready

Document Reader Agent

Extracts facts from bills, PDFs, reports, technical documents and supplier quotes.

Primary handoff

→ Asset Passport Agent→ Finance Modelling Agent→ MRV Agent

Agent 3 Operational Training Pack Ready

Photo / Visual Evidence Agent

Analyses inspection photos and videos while labelling findings as hypotheses until verified.

Primary handoff

→ Asset Passport Agent→ Governance Agent

Agent 4 Operational Training Pack Ready

Asset Passport Agent

Creates the structured digital record of each industrial asset.

Primary handoff

→ Industrial Playbook Matching Agent

Agent 5 Operational Training Pack Ready

Industrial Playbook Matching Agent

Matches assets to modernisation pathways, quick wins and deeper upgrades.

Primary handoff

→ Finance Modelling Agent→ Supplier / Funding Match Agent later

Agent 6 Operational Training Pack Ready

Finance Modelling Agent

Prepares draft CAPEX, OPEX, payback, funding gap and finance-readiness logic.

Primary handoff

→ Governance Agent→ Report Generator Agent

Agent 7 Operational Training Pack Ready

MRV Agent

Separates estimated impact from verified impact and checks baseline/after-data readiness.

Primary handoff

→ Governance Agent→ Public Trust workflows

Agent 8 Operational Training Pack Ready

Governance Agent

Routes high-impact outputs to technical, finance, MRV, safety, privacy and ethical review.

Primary handoff

→ Report Generator Agent

Agent 9 Operational Training Pack Ready

Report Generator Agent

Builds evidence-linked investor memos, board packs, country briefs and public summaries.

Primary handoff

→ Human approval→ External output

Agent 10 Operational Training Pack Ready

Amanah Autopilot Supervisor

Runs overnight checks, finds missing evidence, prepares review queues and generates the morning briefing.

Amanah Autopilot prepares actions for human review. It must not be presented as independently making high-impact decisions.

Agent 11 Operational Training Pack Ready

Waste & Leakage Agent

Detects operational loss, quantifies financial exposure, and keeps actual/estimated/forecast figures visibly separate.

Primary handoff

→ Document Reader Agent→ Finance Modelling Agent

Agent 12 Operational Training Pack Ready

Capital Allocation Agent

Ranks finance-modelled intervention options across 13 dimensions to recommend where the next £1 of capital should go.

Primary handoff

→ Report Generator Agent→ Governance Agent

Capital Allocation Agent produces a governed ranking for human and Council review. It must not be presented as making an autonomous investment decision.

Next-Stage Agents

These 4 agents do not yet have the same full operational training-pack structure as the 12 agents above.

Agent 13 Next Training Pack

Supplier / Funding Match Agent

Agent 14 Next Training Pack

Customer Success Agent

Agent 15 Next Training Pack

Sales CRM Agent

Agent 16 Next Training Pack

Analytics Agent

Council Workflow

Research→ Document Reader→ Photo / Visual Evidence→ Asset Passport→ Playbook Matching→ Finance→ MRV→ Governance→ Report Generator→ Human Approval

Amanah Autopilot Supervisor runs across the workflow

Checks missing evidence
Checks blocked tasks
Checks review queues
Checks readiness
Prepares morning briefing

Agent Council Handoffs

Every handoff should preserve:

Evidence linksMissing dataConfidenceRisk flagsEstimated vs verified statusHuman approval requirementsAudit trail

A downstream agent must never silently increase confidence, drop missing-data warnings or convert estimated data into verified data.

Agent Interdependency Map

Real mechanisms between real agents — not an org chart of imagined connections.

Evidence Verification Active

Research AgentDocument Reader AgentPhoto / Visual Evidence Agent

Cross-checks public context, extracted documents and visual findings against each other before Asset Passport builds a record.

Transition Risk Active

Asset Passport AgentIndustrial Playbook Matching Agent

Surfaces risk_flags (e.g. transition/procurement/baseline risk) that Governance must route for review.

Investor Intelligence Active

Finance Modelling AgentReport Generator Agent

Combines draft finance modelling with the evidence-linked memo an investor or board actually reads.

Funding Match Not yet trained

Supplier / Funding Match Agent

Not yet trained — next-stage agent with no operational training pack. Shown blocked, not simulated.

Agent Council Decision Protocol

Before a high-impact output moves forward:

Evidence exists.
Missing data is visible.
Output schema is valid.
Safety rules pass.
No Harm Gate is checked.
Human approval requirement is evaluated.
Sensitive data is protected.
Estimated vs verified status is preserved.
External action is permissioned.
Audit trail is created.

Council Views

1. Architecture View

Shows all 16 agents and statuses.

2. Operational View

Shows only the 12 agents with full operational training packs.

3. Handoff View

Shows how data moves between agents.

4. Approval View

Shows where human review is mandatory.

5. Evidence View

Shows which agents consume and produce evidence.

6. Overnight View

Shows Amanah Autopilot supervision.

Council Dashboard Cards

Operationally trained agentsNext-stage agentsTotal training filesGolden test casesAgents requiring reviewActive human approval gatesEvidence-producing agentsEvidence-consuming agentsFinance-sensitive agentsMRV-sensitive agentsPublic-output agentsOvernight supervisor status

Example Council Case

Project: Boiler House #3 Modernisation

Research Agent: Finds public/site context.
Document Reader Agent: Reads fuel bills and supplier quote.
Photo / Visual Evidence Agent: Flags visible insulation gaps and soot as hypotheses.
Asset Passport Agent: Builds asset record.
Playbook Matching Agent: Matches Boiler Modernisation Playbook.
Finance Modelling Agent: Drafts CAPEX, OPEX and payback assumptions.
MRV Agent: Checks baseline and after-data requirements.
Governance Agent: Routes technical and finance review.
Report Generator Agent: Builds decision memo.
Amanah Autopilot: Overnight checks missing evidence and prepares morning briefing.
Human: Approves high-impact external use.

Why the Council Matters

EcoIQ should not behave like one black-box chatbot.

The Council model

Separates specialised responsibilities
Preserves evidence traceability
Prevents one agent from doing everything
Makes approvals visible
Creates safer handoffs
Supports enterprise auditability
Makes failures easier to diagnose

Microsoft Ecosystem Integration

EcoIQ AI Agent Council is designed to integrate with:

Microsoft Semantic Kernel conceptsAzure AI Agent Framework conceptsMicrosoft FabricTeams approval workflowsSharePoint evidence packsPower BI operations dashboardsResponsible AI tools

Does not claim

Microsoft certificationMicrosoft partnershipOfficial Microsoft approval

Presentation Mode

12 trained operational agents working as one governed system.

From evidence collection to finance, MRV, governance and reporting — with human approval at high-impact points.

Story 1

Evidence to Asset

Research→ Document Reader→ Visual Evidence→ Asset Passport

Story 2

Asset to Investment

Playbook Matching→ Finance→ MRV→ Governance

Story 3

Decision to Impact

Report Generator→ Human Approval→ Implementation→ Amanah Autopilot Monitoring

Training Pack Repository Layout

Every operational agent's full training pack lives at ai_agents/<agent_name>/ in the repository, and the master index at ai_agents/ai_agent_training_index.md explains execution order, handoffs and demo narratives in full.

ai_agents/amanah_autopilot_supervisor/ai_agents/asset_passport_agent/ai_agents/capital_allocation_agent/ai_agents/document_reader_agent/ai_agents/finance_modelling_agent/ai_agents/governance_agent/ai_agents/industrial_playbook_matching_agent/ai_agents/mrv_agent/ai_agents/photo_visual_evidence_agent/ai_agents/report_generator_agent/ai_agents/research_agent/ai_agents/waste_leakage_agent/

Safety and Governance

EcoIQ AI agents are specialised decision-support workflows, not fully autonomous decision-makers.
Twelve agents currently have full operational training packs in the repository.
Four additional agents are next-stage and do not yet have the same full training-pack structure.
High-impact industrial, financial, MRV, public reporting and Islamic finance outputs require human review.
Visual findings remain hypotheses until verified.
Estimated impact must remain separate from verified impact.
Maqasid/Mizan is ethical decision-support, not a fatwa.
Microsoft ecosystem-ready does not mean Microsoft certified or Microsoft partner.