Kava Agent · Kava OS Autopilot
It drafts the day. You decide how much it decides.
Kava Agent is an AI agent for vending operations. It reads telemetry, forecasts, prices, service tickets and routes, then does the routine work: pick lists, route changes, markdowns, service dispatch and supplier orders. Every action type runs at the level you choose: Suggest, Approve or Auto.
- of observation before it proposes autonomous actions
- 14 daysof observation before it proposes autonomous actions
- control levels: Suggest, Approve, Auto
- 3control levels: Suggest, Approve, Auto
- rollback for price changes
- 1 clickrollback for price changes
Six jobs it takes off your morning
Pick lists
Drafts and publishes pre-kitting pick lists per machine and per spiral, ready before the warehouse opens.
Routes
Re-sequences today's route around a device that went offline, a compressor alert or a host's event.
Prices
Proposes and applies price changes within your guardrails, such as a maximum ±15% move and never below cost.
Service tickets
Opens, prioritizes and dispatches tickets with the temperature curve, door log and sales impact attached.
Supplier orders
Drafts restock orders to suppliers from forecast demand across the whole fleet.
Plain-English answers
"Which coolers in Building C ran warm this week, and did we lose sales?" Ask the way you'd ask a colleague.
Suggest, Approve, Auto. Set per action, not per agent.
Trust isn't all or nothing, so the setting isn't either. Each action type gets its own level. You might let the agent publish pick lists on its own, ask before it changes a route, and only suggest prices until you've watched it for a month.
Auto never means unlimited. It means the agent acts inside the guardrails you wrote, such as "never below cost" or "max ±15%", and tells you afterwards. Every action lands in the audit log with its inputs, its setting and who approved it. Every price change can be rolled back with one click.
- Suggest: it drafts the action; a person carries it out.
- Approve: the action is ready; one tap sends it.
- Auto: it acts within guardrails and reports what it did.
What changes when software acts instead of answering
| Situation | Chat assistant | Kava Agent |
|---|---|---|
| Tomorrow's pick lists | Explains how to build one | Drafts them overnight and publishes at your setting |
| A cooler is running warm | Answers if someone asks | Opens the ticket and moves the stop up the route |
| Sandwiches expire tonight | Suggests a discount | Applies a markdown within your guardrails, logged and reversible |
| Next week's cold brew order | Lists what sold last week | Drafts the restock order to the supplier |
| Accountability | A chat history | An audit log of every action and approver |
An example rollout
- Days 1–14
Watch and learn
The agent observes telemetry, sales, routes and the decisions your team makes. It answers questions but proposes no autonomous actions.
- Day 15
First drafts
Proposals start in Suggest: tomorrow's pick lists, a re-sequenced route, a markdown on items dated tonight.
- Weeks 3–4
Move routine work to Approve
Once its pick lists match what you would have built, switch them to Approve. Each one is a single tap.
- Month 2 on
Auto where it's earned
For example, let pick lists publish automatically and keep prices on Approve. Any action can move back down at any time.
What a 2 a.m. compressor alert looks like with an agent
Signal
A cooler at a distribution center shows long compressor run cycles and slow temperature recovery.
Plan
The agent opens a service ticket, moves the site earlier on today's route and drafts a markdown for the fresh items inside.
Ask
Route changes are set to Approve, so the ops manager gets one notification with the change and its reasoning.
Act and log
The driver app updates, the technician is dispatched, and the audit log records the inputs, the action and the approver.
Undo if needed
The fix holds and the markdown wasn't needed after all. One click restores the previous prices.
The agent drafts tomorrow's pick lists before my drivers wake up. I approve them over coffee.
Guardrails and accountability
Guardrails
- Price moves
- A maximum per change, for example ±15%
- Price floor
- For example, never below cost
- Scope
- Per item, category, site or action type
Accountability
- Audit log
- Every action with its inputs, setting and approver
- Rollback
- One click for price changes
- Who sets levels
- Owner and manager roles
Plan
- Included in
- Autopilot and Enterprise
- Autopilot price
- $69 per device per month; $58.65 with annual billing
What would you stop doing if you trusted the draft?
Most ops managers spend the first hour of the day rebuilding a plan that is mostly yesterday's plan with a few changes. The agent's job is to find those changes. Yours is to decide whether they're right. Write down three decisions you make every morning that you could defend in your sleep. Those are your first candidates for Approve, and later for Auto.
Kava Agent questions
Can the agent change prices without my approval?
Only if you set price changes to Auto, and only within your guardrails, such as a maximum ±15% move and never below cost. Every change is logged and can be rolled back with one click. Dynamic Pricing explains the rules it works inside.
What data does it use?
Everything Kava OS already holds: telemetry from Fleet Command, forecasts, prices, service tickets and routes. It doesn't need a separate data feed or a new integration to start.
What happens during the 14-day learning period?
It observes your fleet and your team's decisions and answers questions, but proposes no autonomous actions. After 14 days, proposals begin in whatever mode you've set for each action type.
Who is accountable when it acts on Auto?
You set the level and the guardrails, and the audit log shows exactly which setting allowed each action. Owners and managers can move any action type back to Approve or Suggest at any time.
Which plan includes Kava Agent?
Autopilot, at $69 per device per month or $58.65 with annual billing. It builds on Forecasting & Replenishment and Dynamic Pricing, which are included.
Where to go next
Let it draft tomorrow. You still sign off.
Book a demo and we'll show you a day of agent drafts on a fleet like yours, from the pick lists to the audit log.