OPS-243
Check eligibility coverage
Review the next eligibility batch and group records that need operator attention.
Front Range Health
Eligibility Check
TO RUN
AI Operations
AI Operations is the runtime where skills and agents move from a successful test into routine work. It protects against AI harm, AI cost, and HIPAA exposure with logging, monitoring, human checkpoints, undo, and an emergency stop. AI Operations runs on Nucleus, the governed foundation that keeps connected data, typed tools, and every execution inside the perimeter.
What the runtime gives you
AI Operations is where a tested skill or agent starts doing routine work. There is no platform fee: you pay for the AI usage the work consumes, and nothing else.
Monitoring you can read at a glance
Every running session reports its state, its elapsed time, the tokens it has consumed, and what that costs, so an operator sees where attention is needed without opening a log.
Sign-off on every write
Agents propose, a named human approves. A binding change waits at the gate until somebody with the authority to release it does.
Cost control before the bill
Token meters and per-task cost bars make a runaway process visible while it is running, not at the end of the month.
Emergency stop and undo
Halt active work from one control, and revert an approved action that still produced the wrong result.
Run the work without losing control
A seeded payer operation on an ordinary afternoon: every task stays visible, token and cost budgets stay in view, write actions wait at a named human approval gate, and operators retain emergency stop and undo.
Seeded demo data 2 agents working
Agent workboard
Follow payer work from intake to completion. Every write pauses at a named human gate.
Human approval Emergency stop ready Undo recorded
Tasks on the board
7
Human decisions
2
Tokens used today
105.2k
Seeded run cost
$6.10
Today's payer work
Synthetic NEMT examples with no member or trip-level details.
Ready
OPS-244
Watch credential expirations
Monitor vehicle and driver credential windows without changing source records.
Bluebird Mobility
Credential Monitor
In progress
OPS-241
Sweep mileage underpayments
Compare CO-45 remits paid at $1.10 per mile with the $1.25 fee schedule.
Summit Care Partners
Claims Recovery
OPS-242
Measure slow-pay drift
Track remit aging as it grows 3 days every 10 days toward the 60-day cap.
Pinnacle Rides
Payment Watch
Human review
OPS-245Finance lead gate
Submit correction batch
The March 15 to April 15 mileage correction batch is ready for release.
Summit Care Partners
Claims Recovery
OPS-246Revenue manager gate
Open slow-pay alert
The projected remit lag now meets the operator-defined alert threshold.
Pinnacle Rides
Payment Watch
Done
OPS-240
Prepare claims follow-up
A claims batch summary is ready for the next operator review.
Front Range Health
Eligibility Check
Named human gates
Agents can analyze and draft. A person must release every write shown here.
- Finance lead gate Demo only
Submit mileage corrections
Release the Summit Care Partners correction batch for the seeded recovery window.
Proposed by Claims Recovery
- Revenue manager gate Demo only
Open collections alert
Notify the finance team before Pinnacle Rides reaches the projected lag cap.
Proposed by Payment Watch
These controls change this seeded page only. They never write to an operational system.
Runs on real frontier models, in your own cloud
AI Operations runs Claude Code and Gemini models on Google Cloud's Gemini Enterprise Agent Platform (formerly Vertex AI), inside the same HIPAA-eligible perimeter as your data, metered and gated by the runtime above.
Claude Code
Gemini
Google Cloud
How governed work is classified
The runtime uses the same four data classes shown in the Nucleus refinery, so protection language stays consistent from connected data to active work.
Scripted, deterministic
Rules check schemas, keys, dates, and units before AI touches the work.
Structured, AI checked
Structured records can be tagged and checked against known fields and constraints.
Unstructured, human checked
Documents, scans, and free-form material pass through a human checkpoint before anything becomes binding.
Sensitive, sealed
Sensitive content stays marked and sealed so agents receive only the minimum-necessary context.
Everything else the runtime does
The rest of the runtime behind the workboard.
Browser-control agents in governed VMs
Watch an agent drive a real web app from a controlled VM with a live view, working the portals and back-office tools your team uses today.
From exploration to deterministic automation
Agents turn what they learn at the browser into replayable Playwright automations, so routine runs become fast, cheap, and predictable.
Claude Code on Google Cloud
Frontier coding agents run on the Gemini Enterprise Agent Platform (formerly Vertex AI) inside the HIPAA-eligible perimeter, under the same Business Associate chain as the rest of the platform.
Protection from AI harm
Typed, scoped tools and human approval before binding changes keep model uncertainty from turning directly into action.
Protection from AI cost
Monitoring shows what is running and gives operators a way to stop work before a runaway process becomes a runaway bill.
Protection from HIPAA exposure
Minimum-necessary, tenant-scoped access limits each agent to the context its task requires.
Emergency stop
Halt active work from one control when a workflow behaves unexpectedly or its operating conditions change.
Undo
Versioned, revertible changes give operators a path back when an approved action still produces the wrong result.
Logging and monitoring
See what ran, which tools it called, what context it received, and where a human approved the work.
Run with control
AI Operations is coming soon. A demo video of a browser-control agent moving bookings from a brokerage portal into a live dispatch demo is on the way.