High-voltage SME sites
- Audience
- Contracted capacity ≥100kW, equipped (BMS comms + metering foundation)
- Includes
- All five Agents — ESS dispatch, natural-language diagnosis, AI sustainability reports, AI load forecasting, multimodal O&M
Products
How each Agent works,
and how to onboard
The home page gives the five-Agent overview. Here we detail each Agent's inputs, decision logic, outputs, and use cases, plus the full onboarding flow from consultation to Agentic upgrade. The middleware's protocol integration and second-level monitoring are covered on the home page and not repeated here.
Product lines
Plans matched to your site scale
lvsota Agentic AI focuses on high-voltage SME sites and export-heavy heavy industry; basic carbon accounting for micro sites is handled by our affiliate Shihe and does not carry the Agentic AI brand.
Heavy-industry exporters
- Audience
- Steel / aluminum / cement / fertilizer / hydrogen and other export-heavy industry
- Includes
- Full L2 + CBAM module (roadmap) + multi-site comparison + multilingual manuals
Agent capability depth
How the five Agents work in detail
ESS dispatch
- Input
- Tariff forecast · load forecast · battery SOC/SOH · contracted capacity
- Decision logic
- Numerical-layer MPC/MILP solves the charge/discharge schedule, with LLMs handling strategy adaptation, anomaly response, and explainable decisions — balancing revenue, battery life, and overrun
- Output
- ESS charge/discharge schedule (per-interval commands) + schedule rationale card
- Use case
- Storage sites, peak shaving & valley filling, tariff arbitrage
- Limits
- Constrained by battery hardware capability and safety boundaries; the rationale card extracts structured reasons from solver decision variables, then LLM-polished to avoid hallucination; if the cloud LLM API drops, the edge continues on the last approved weights and flags degraded mode
Natural-language diagnosis
- Input
- Natural-language question (e.g., why did we exceed contracted capacity?) + site-wide energy data
- Decision logic
- Rule engine filters candidate causes + LLM ranks by causal templates + confidence tagging, correlating consumption, contract, device state, and events
- Output
- Plain-language explanation + recommended action
- Use case
- Quick manager Q&A, anomaly triage, report interpretation
- Limits
- Depends on data completeness — missing data is stated, not guessed; typical response 30–90s; four-period contracts compute overrun per period; multi-site comparison requires normalized production schedules
AI sustainability reports
- Input
- Energy data · emission factors · ESG indicator inputs
- Decision logic
- A deterministic calculation layer computes emissions, while the LLM maps across standards (written in after human review) and drafts sections
- Output
- Sustainability report draft + inventory working papers (for accountant assurance)
- Use case
- Listed-company sustainability reports, ESG disclosure; CBAM module as an add-on for export-heavy industry (roadmap)
- Limits
- No attestation (drafts go to the accountant); IFRS S1/S2 apply in three phases 2026/2027/2028; Taiwan grid emission factor 0.474 kg CO2e/kWh for 2024; CBAM defaults to the average of the top-10 highest-emitting countries (with uplift, per the current EU regulation); v1 ships Scope 1+2, Scope 3 optional
AI load forecasting
- Input
- Historical consumption · weather · schedules · external factors
- Decision logic
- Time-series modeling + external-factor correlation, with the LLM handling orchestration, explanation, and degradation
- Output
- Next-day load forecast + basis for contracted-capacity adjustment decisions
- Use case
- Provides next-day decision basis for cutting overrun and idle demand charges
- Limits
- High-voltage time-of-use tariff summer months 6/1–9/30; MAPE in literature 1.6–8% at distribution level, 8–13% at site level, up to 59–74% for industrial processes — actuals verified by site backtest; most SMEs rely on manually annotated startup plans, MES/ERP is a plus; initial warm-up required
Multimodal O&M assistant
- Input
- Voice · text · image (on-site photo)
- Decision logic
- VLM reads meters / nameplates / error codes + ASR voice + equipment-manual RAG retrieval, with safety-graded routing (high-voltage sections give observation-only advice)
- Output
- On-site repair guidance steps + safety-grade labeling
- Use case
- On-site O&M, fault triage, onboarding new staff
- Limits
- Does not replace a licensed electrician (arranged via ESCO association); edge hardware around TWD 80–200k; CMMS write-back is a phase-2 option, defaults to lightweight export; manuals initially Chinese and English; odor / noise go through the voice path, not images; undocumented equipment returns generic advice
Onboarding flow
From consultation to Agentic upgrade
Consult & assess
Understand your device protocols, site scale, pain points, and goals — confirm the feasible scope.
TBD · weeksPOC trial
The middleware connects to existing equipment for validation; second-level monitoring lands first, then we review data quality and site feasibility.
TBD · weeksBasic deployment
The middleware goes into production: protocol integration, second-level monitoring, and contracted-capacity visibility go live.
TBD · weeksAgentic upgrade
Enable the five Agents on demand, moving from monitoring to autonomous decisions. Prerequisites: BMS protocol support, smart meter / CT metering, ≥30 days of historical data — pass these before L2 assessment.
On demandRequired resources
- Device protocol list TBD · on-site device models and protocols (Modbus/MQTT/REST)
- Network Site needs stable intranet/internet for middleware and Agent comms
- Onboarding staff TBD · site-side personnel needed (energy management / IT / O&M)
Plan your onboarding flow
Tell us your device protocols and site scale — we'll arrange a POC start and upgrade cadence matched to your maturity.