AimoAssist

A purchasing agent for professional kitchens

Trustedtrade agents

A kitchen's AI agent orders food from Valio Aimo. Here is what it is told, what it decides, and what it does, for one real order.

Open the customer appValio Aimo's dashboard

Agent run · hotel-aurora · Tue 17 NovReady0.0 s
Kitchenmenu · recipes · habitsLanguage modelreads the requestValio Aimocatalog · prices · stockDecision modelis this the product?Risk & scoringon time · price · fitRulesthe signed slipHead chefa personWaiting for a request
› idle

How it decides

Think, call, see,then decide

For every line of the cart the agent runs the same loop: it asks itself a question, calls the one tool that can answer it, reads the answer, and only then decides. Pick a line to watch it work.

AimoAssist agent · crème fraîche · step 1 of 5ReadyWaiting to start
  1. 01

    ThinksWhat does this kitchen usually buy?

    CallsKnowledge base
    product_memory("creme fraiche, vegan")

    Sees…

  2. 02

    ThinksCan I get it for Tuesday?

    CallsGateway
    POST /v1/quotes { 469175, "2026-11-17" }

    Sees…

  3. 03

    ThinksWhat can stand in for it?

    CallsGateway
    GET /v1/products/469175/substitutes

    Sees…

  4. 04

    ThinksIs it really a crème fraîche?

    CallsDecision model
    yes_no("Is Oatly iMat Fraiche a crème fraîche?")

    Sees…

  5. 05

    ThinksDoes it keep the vegan guests safe?

    CallsRules
    evaluate(slip, line) · R05b

    Sees…

01Ordering

By hand,then by agents

A chef opens Valio Aimo's 25,000 products one page at a time. Agents check them all at once, guided by the kitchen's knowledge base: its dishes, diet rules and usual products.

They find a substitute when stock runs out, and when the chef changes a line they learn, once replaying past carts breaks no rule.

  1. 01By hand: one page at a time
  2. 02The knowledge base guides the agents
  3. 03The swarm searches the whole catalog
  4. 04A product runs out: a substitute
  5. 05The head chef corrects a line
  6. 06The agents note it and learn

02The trust protocol

Signed,not trusted

The agent never touches Valio Aimo's systems. It talks to a gateway that answers with signed offers: price, stock and delivery date, valid for 15 minutes.

Anyone can check the signature with the published key. Change one digit and it fails.

03Guardrails and approval

Only a humansigns

The kitchen manager signs a permission slip: what the agent may spend, which diet rules hold, and who approves the rest. When a line needs a person, the head chef signs, and the signature covers those exact offers and quantities, nothing else.

The gateway checks every signature again. A rogue agent that tries to cheat is stopped, with the rule that caught it.

04Exceptions

When stockruns out

After the order is placed, the agent keeps checking it until the cut-off.

If a product runs out, it proposes a substitute that keeps the diet rules, for the chef to approve.

  1. 01The order is confirmed
  2. 02The agent watches it before the cut-off
  3. 03A product runs out
  4. 04It proposes a replacement

05Learning

It learns,carefully

The agent remembers the kitchen's habits. Three swaps make a product preferred, and the amounts follow what the chef keeps editing.

A change goes live only if replaying recent carts shows it breaks no rule. It never changes the slip.

06Openness

Every stepon record

Each step lands in an audit trail that seals the step before it, in sentences the kitchen and Valio Aimo can both read.

Any agent can plug in: the interface is built on open, published standards, not one platform's format.

ResultFrom our own test runs

Fast, andinside the rules

30/30
scripted orders matched their expected outcome
10/10
planned attacks blocked, plus 19/19 red-team
45/45
right product first in the match eval
6–7 s
from a request to a 20-line cart, live

Try itThree ways in

See it work,from your side