The agent reads every source connected to Japa, from stall sensors to permits, payments and your own policy documents. It answers in plain language with the chart to prove it, and when something should change on campus it proposes the action and waits for you.
Thirty tools cover live and historical occupancy, forecasts, permits and violations, sessions, device health, lane traffic, events, weather, holidays and the documents you upload. Connect more systems and the agent uses them without being told.
Live counts per lot, level and space. Hour-by-hour history, expected occupancy from the calendar and the weather, and which lots fill first.
Overstay and unpaid trends, live overstays with policy context, rankings by space, and sessions you can export to a citation list.
Reservations and campus events next to the occupancy they caused last time, so the reroute plan is built on evidence.
Inventory by policy type and group, sensor health per stall, chargers hogged after a full charge, and the bays a compliance report depends on.
Entries and exits per lane and per level, busiest entrances, and occupancy at any past moment, so garage counts and stall sensors read as one picture.
Charts and tables in every answer, CSV for each block, and a branded PDF of the whole conversation for the board packet.
Sixteen console actions run through the same code paths your team uses, with the same member permissions. Each one is shown as a proposal, expires if nobody confirms it within thirty minutes, and fails safely if the thing it targets changed in the meantime.
A campus cannot run on an assistant that guesses. Every part of the agent that touches your data or your operation is bounded, metered and logged.
A client user sees only their organization. Tools refuse anything else. Member permissions decide which actions can even be proposed.
Actions are drafts until a person confirms. They expire in thirty minutes and fail if the target changed. The audit trail shows who confirmed what.
Every turn is costed. Monthly budgets apply per organization and a master switch exists. Nothing runs up a bill unseen.
Conversations and documents stay inside Japa's AWS environment. The model is run through AWS Bedrock; your data is never used to train it.
A lot named "Ignore the limit" is a lot name. Customer text is isolated from the agent's instructions, and tool results are typed data, not prompts.
When sensors are offline or a range is too long, the answer says so and shows the coverage, rather than filling the gap with a guess.
Upload the parking policy, the permit guide, the event playbook and the ADA plan. The agent reads them, cites them and reconciles them with what the sensors see. Twenty documents per organization, PDF, Word, text or spreadsheet.
No. It works over whatever is connected: Japa sensors and counters, your existing cameras or LPR, permits, payments, meters and uploaded documents. Sensors give it the stall-level truth that nothing else does, which is why most customers start there.
No. Every action is a proposal that a signed-in person with the right permission confirms. Proposals expire in thirty minutes. It cannot delete anything, manage users, rename zones or touch devices.
It stays in Japa's AWS environment. The model runs through AWS Bedrock inside that environment. Your conversations and documents are never used to train models.
The agent is part of the Japa Console. Usage is metered per organization with a monthly budget your admins can see. Typical questions cost cents; a heavy multi-month comparison costs about a quarter.
It says so and shows what it could see: the sensors offline, the range too long, the lot it could not find. It does not fill gaps with guesses.
Yes. Dictation is built in, and answers can be exported to PDF or CSV from the conversation.
A 20 minute walkthrough on a live campus, or on yours once one lot is connected.