How prachit27/gridmindai works
GridMind accepts a workload through its public dashboard or negotiation API, then coordinates four independent specialists. Their verdicts are checked for joint physical consistency; if they conflict, the orchestrator sends peer positions for another bounded round and returns an auditable decision or escalation. Specialists use a shared harness to assemble scoped facility data, external signals, and policy context, then validate model verdicts with domain guardrails. The README documents memory precedents, decision history, PDF reports, and isolated domain databases. Some subsystem wiring is based on README documentation because its source was not sampled.
Main entry points: orchestrator.py Negotiation Loop · harness.py Shared Agent Harness · gateway/main.py Agent Gateway · web/main.py Web API
Components 22
Negotiation
- Negotiation API
agents/orchestrator/main.py - Negotiation Loop
agents/orchestrator/orchestrator.py - Consistency Check
agents/orchestrator/consistency.py - Decision and Escalation
agents/orchestrator/decision.py - Decision Report
agents/orchestrator/report.py - Agent Gateway
agents/gateway/main.py
Specialists
- Power Specialist
agents/power/agent.py - Cooling Specialist
agents/cooling/agent.py - Facilities Specialist
agents/facilities/agent.py - Cost Specialist
agents/cost/agent.py
Agent Harness
- Shared Agent Harness
agents/common/harness.py - Constraint Context
agents/common/constraint_context.py - Domain Guardrails
agents/power/guardrail.py
Data and Integrations
- Gemini Model
- Facility Model
seed/facility_model.py - External Signals
agents/common/external_signals.py - Policy Context
agents/common/policy.py - Firestore Databases
Other components
- Workload Operator
- Grid, Weather, Operations
Connections 35
- Workload Operator → Dashboard: uses
- Dashboard → Web API: submits workload
- Web API → Negotiation API: forwards negotiation
- Web API → Firestore Databases: reads history
- Negotiation API → Firestore Databases: loads workload
- Negotiation API → Negotiation Loop: starts negotiation
- Negotiation Loop → Agent Gateway: routes specialist calls
- Negotiation Loop → Power Specialist: runs locally
- Negotiation Loop → Cooling Specialist: runs locally
- Negotiation Loop → Facilities Specialist: runs locally
- Negotiation Loop → Cost Specialist: runs locally
- Agent Gateway → Power Specialist: dispatches
- Agent Gateway → Cooling Specialist: dispatches
- Agent Gateway → Facilities Specialist: dispatches
- Agent Gateway → Cost Specialist: dispatches
- Power Specialist → Shared Agent Harness: invokes
- Cooling Specialist → Shared Agent Harness: invokes
- Facilities Specialist → Shared Agent Harness: invokes
- Cost Specialist → Shared Agent Harness: invokes
- Shared Agent Harness → Constraint Context: assembles
- Constraint Context → Firestore Databases: reads scoped data
- Constraint Context → External Signals: incorporates
- Constraint Context → Policy Context: incorporates
- External Signals → Grid, Weather, Operations: obtains conditions
- Facility Model → Firestore Databases: seeds facility data
- Shared Agent Harness → Gemini Model: requests verdict
- Shared Agent Harness → Domain Guardrails: validates verdict
- Domain Guardrails → Power Specialist: returns checked verdict
- Negotiation Loop → Consistency Check: checks round
- Consistency Check → Negotiation Loop: returns conflicts
- Negotiation Loop → Decision and Escalation: resolves or escalates
- Negotiation Loop → Firestore Databases: recalls precedents
- Negotiation Loop → Firestore Databases: records decision
- Negotiation API → Decision Report: renders result
- Web API → Decision Report: renders PDF