AI Agents

SETI AI ships with a small, purpose-built team of AI agents — Bob, Bill, and Mini — that work together to keep your energy storage fleet safe, understood, and optimized. They are not generic assistants. Each one has a defined role, a hard-coded knowledge of SETI Power hardware, and a place in an autonomous operations loop. Every answer they give is grounded in your knowledge base and cites its sources.

Related pages: Command Center · Agent Tools · Notifications

The team at a glance

The three are designed to mirror a real operations crew: a field technician who watches the equipment (Mini), an analyst who investigates what the data means (Bill), and an operations lead who makes the call and acts (Bob).

Shared foundation — what every agent knows

All agents are built on the same non-negotiable knowledge of SETI Power hardware, so they never apply generic lithium-ion assumptions to your graphene supercapacitor packs:

• Cell: SETI WT-4.2V21000F graphene supercapacitor (electrostatic storage, not lithium) — 3.8 V nominal, 4.2 V max, 3.0 V min, 21,000 F, 30,000+ cycle life, 30-year design life.
• 52 V pack (14S-5P): 58.8 V max, 100 Ah / 5.2 kWh, 200 A continuous, JK-BMS over BLE.
• 24 V pack (7S-5P): 29.4 V max, 100 Ah / 2.6 kWh, 100 A continuous.
• Hard safety limits: emergency temp disconnect 60°C, warning 55°C, cell-balance alarm 50 mV, balance trip 100 mV, overvoltage cutoff 4.25 V/cell, undervoltage cutoff 2.8 V/cell.

Because these specs are baked in, the agents catch issues a lithium-trained assistant would miss — and they understand why your packs behave differently (no thermal runaway from chemical decomposition, far higher cycle life, faster charge/discharge, but real electrolyte degradation above 60°C).

Grounded and cited. Every agent has live access to your Knowledge Base (uploaded PDFs, manuals, datasheets, internal specs). They are instructed to search it first before answering and to cite the source filename and section in their reply — and to say so honestly when the knowledge base has no answer, rather than inventing one.

Bob — Senior Operations Lead

"Approved." / "Rejected — reason." / "Hold for human review — reason." Safety first, always.

Role. Bob is the decision-maker, command approver, and safety enforcer for your fleet. He is the agent you give an instruction to and the one who says yes, no, or "escalate to a human."

Personality & purpose. Decisive, operational, and authoritative when safety is on the line; relaxed and conversational for everyday questions. Bob leads with the verdict, then the reasoning, and always cites the specific safety threshold behind a decision. His guiding rule: when in doubt, reject and escalate.

What Bob does for the operator:

• Approves or rejects proposed actions, validated against your hard safety limits.
• Issues commands to packs, BMS units, and inverters (relay control, load shed, BMS reset).
• Runs, stops, and monitors visual workflows you've built in the Node Editor.
• Reports fleet status and pack status on demand.
• Maintains the operational audit trail — a logged record of every command decision.
• Notifies you (and escalates) when human judgment is required.

Before approving anything, Bob runs his own safety check: thresholds within limits, a clear shutdown path exists, logging and notification steps are wired, and the action is scoped to the correct user. If any check fails, he rejects and explains why.

Bill — Research Specialist

Lead with the key finding. Summary → detail → recommendation.

Role. Bill is the intelligence analyst for your energy storage systems. He turns raw BMS data — cell voltages, temperatures, state of charge, state of health, balancing status — into clear, actionable intelligence.

Personality & purpose. Concise and analytical. Bill leads with the headline finding, then supports it with structured detail and a recommendation. He matches a casual tone for casual questions and goes deep when the question is technical.

What Bill does for the operator:

• Analyzes degradation patterns, thermal profiles, and SOH trends across packs.
• Detects anomalies statistically and follows up automatically with root-cause analysis and cross-metric correlation.
• Forecasts degradation — and flags immediately if a pack is projected to drop below 80% health within six months.
• Produces full health reports and technical write-ups for customers.
• Recommends and builds monitoring pipelines: when you ask about a sensor, Bill researches its specs (protocol, address, range, accuracy), verifies them against pack limits, and assembles the correct Node Editor pipeline for you.
• Leans heavily on the knowledge base — it is his first tool for any question touching a product, datasheet, or past report, and he always cites the source.

Bill packages his findings as a Research Package — analysis plus a recommended action plus a confidence level — and hands it to Bob for approval. He does not act on the field himself; he investigates and advises.

Mini — Edge AI Field Agent

Role. Mini is the agent that lives on the hardware. Running on an edge AI device (Jetson) next to the battery pack, Mini watches the Battery Management System in real time and is the first to notice when something is wrong.

Purpose. Mini is the team's eyes in the field. Where Bob and Bill reason from the dashboard, Mini reacts at the edge — continuously monitoring live cell data and raising anomaly alerts the instant a metric drifts out of bounds, without waiting on a cloud round-trip.

What Mini does for the operator:

• Real-time BMS monitoring directly on the equipment.
• Continuous anomaly detection on live telemetry.
• Publishes alerts (overvoltage, thermal, anomaly, and more) onto the system's MQTT message bus, kicking off the autonomous response loop.

Mini is distributed as a dedicated, compact AI model (morwic/seti-mini) tuned to run efficiently on edge hardware — small enough for the field, smart enough to know when to raise its hand.

How they collaborate — the autonomous loop

The three agents form a closed operations loop, coordinated over an internal message bus and an inter-agent notification system:

Mini  →  MQTT  →  Bill (analyze)  →  Bob (decide)  →  Action  →  Notify operator
(edge alert)      (research pkg)     (approve/reject)  (command)   (email/Discord/Slack)

1. Mini detects an anomaly on the live pack and publishes an alert.
2. Bill picks it up, pulls the pack's history, SOH, and thermal profile, runs anomaly and root-cause analysis, and produces a Research Package with a recommendation and confidence level.
3. Bob validates that recommendation against the hard safety limits and approves, modifies, or rejects it — logging the decision to the audit trail.
4. If the action is approved and safe, it is issued to the field.
5. For any critical or warning-level event, the system fires external Notifications so you hear about it even when you're not watching the dashboard.

Inter-agent notifications. Bob and Bill can message each other directly with a priority level — info, warning, or critical. Bill flags Bob the moment he finds something that needs a decision; Bob asks Bill for research before approving; each checks an inbox at the start of every conversation. A critical message demands immediate attention. This is how the team stays coordinated without a human relaying messages. (See Agent Tools for the underlying notify_bob, notify_bill, and check_inbox tools.)

Why this matters for you

• Always-on coverage. Mini watches around the clock; you don't have to.
• Analysis you can read. Bill translates BMS noise into a finding and a recommendation.
• A safety gate that never sleeps. Bob checks every action against your hard limits before anything happens to the hardware.
• Hardware-correct from the start. The team understands graphene supercapacitor behavior — no dangerous lithium assumptions.
• Trustworthy answers. Every response is grounded in your documents and cites its sources.