Written in your voice, consistent with everything the demo claims. Green boxes are the one-line versions to fall back on under pressure. Amber boxes are facts only you can fill in, decide them before the event so you never improvise a commitment on stage.
Your model runs in its own isolated environment; nothing is shared or co-mingled across customers. Where it physically sits depends on your requirements.
No. The LLMs are not trained on your data, and that's the whole point of the model: it hands the AI your context at question time instead of baking your business into anyone's weights.
The next sync carries the change, because permission is stamped on the data as it comes in, every time it comes in. The model mirrors the source system's answer to "who can see this", it never invents its own.
The model shows the truth of your systems, including the ugly parts, and puts it next to what your people know. That contrast is the fastest bad-data detector there is. Wrong data hiding in a system nobody looks at stays wrong forever; wrong data on the model gets caught.
Three things. It reasons over the model, not from memory, so it's grounded in your actual objects and numbers. For the numeric parts it calls deterministic models, not the LLM. And nothing it produces goes anywhere without a person sending it, so a bad draft costs you a review, not a shipment.
No. Agents draft, people send. Nothing reaches a client and nothing posts to a ledger without a human sending it, and that's a term in the customer contract, not a setting someone can flip.
You reject it, and the rejection is signal we use. But in practice the drafts start from your data and your own planners' rules of thumb, so they begin where your best person would begin, not from a blank page.
We're deliberately model-agnostic. The model layer, the part that matters, is ours; the LLM behind it is swappable. We use frontier models where the reasoning is hard and cheaper ones where it isn't.
The connectors do the heavy lifting on data, and the knowledge sessions run in parallel with your people, so the two tracks don't queue behind each other.
No, and you shouldn't. Your systems stay the systems of record and the systems of action. Orchid sits on top: it reads from them, and it drafts back into them. The day you turn Orchid on, nothing about how your business runs has to change.
Messy is the normal case, clean is the exception. The collectors normalise what's there into the four primitives, and the knowledge capture fills the gaps your systems never held. Honestly, if your data were perfect you'd need us less.
Access to the systems you already run, and a few hours with the people who actually make the decisions. That's genuinely the list.
It's the model watching something over time: a stock level, a lead time, a risk. Entities are the things, knowledge is how your people think about them, trackers are what's changing, and actions are what you can do about it. Four primitives, no fifth.
They're structured interviews with the people who run the work, turned into assertions in the model, and every assertion is anchored back to its evidence, you can always see who said it and why. When two people disagree, both views go in with their anchors. That disagreement is usually the most valuable thing we surface all month.
A warehouse gives the AI tables. The model gives it your business: the objects, the relationships, your people's rules, and the actions it's allowed to draft. And the writeback path with a human on send is the part no copilot gives you, a copilot answers questions, Orchid moves work.
We share one conviction with them: the model of the business is where the value is. We bet differently on how you build it. We treat what's in your people's heads as first-class data, we ship agents as configuration rather than consultant-built workflows, and we put "people send" in the contract. And we're building it for companies your size, without an army on site.
Because the knowledge capture is hands-on, and we'd rather make five models excellent than fifty shallow. The packs that come out of those five are exactly what makes the next fifty fast.