Platform
Your data. Your model. Your terms.
Every customer gets their own analyst, running on open weight models we host. No token meter. No third-party AI vendor holding your numbers. And if you want to own the model outright, you can.
You can see it. You can keep it. You can leave with it.
Most AI products rent you access to a closed model owned by someone else. Their price, their terms, their deprecation schedule, their data policy, and their decision to change any of it on a Tuesday. Building a farm's risk management on that is a bet on a vendor's goodwill.
We run open weight models instead. The weights are inspectable, portable, and durable. What runs for you today can still run for you in five years, on hardware you choose, whether or not we are in the room.
Proof
This is not a roadmap slide. We already run a self-hosted Mistral model on our own GPU, serving an on-call ag economist against a governed, read-only warehouse, 24 hours a day. The pattern in the tiers below is the pattern we run ourselves.
A shared model never means shared data.
Your agent is yours. Its own context, its own memory, its own governed data scope. It sees two things: the market warehouse, which is identical for every customer and contains no one's private information, and your data, which is only ever yours.
The model is the engine. It is not the filing cabinet. Isolation happens at the agent and data layer, which is why we can host one model for everyone without anyone's numbers touching anyone else's.
Ask the follow-up question.
Per-token pricing punishes curiosity. It makes you ration the third question, which is usually the one worth asking. We host the model on capacity we own, so the cost is a machine, not a meter. Ask all day.
Pick your level of control. The product works the same at every rung.
| Option | Where the model runs | Where your data lives | Who owns the weights | For |
|---|---|---|---|---|
| Shared desk agent | Our GPUs, our VPC | Our warehouse, isolated per customer | Open weights, we operate | Most customers. The full product, no token meter. |
| Dedicated instance | Our GPUs, capacity reserved for you | Our warehouse, isolated per customer | Open weights, we operate | Heavy use, latency guarantees, stricter isolation. |
| In your cloud | Your account, your VPC | Never leaves your environment | Open weights, you operate | Security teams who need the data to never cross the boundary. |
| Own your model | Anywhere you want | Yours | Yours, exportable | Anyone who wants a model tuned on their own operation and intends to keep it. |
| Bring your own model. | Your existing endpoint | Per your setup | Yours | Teams already standardized on a model or provider. |
Under the table: Move up or down the ladder without changing how anything works. The app, the API, and the MCP server are identical at every rung.
Trained on your operation. Handed to you.
Fine-tune on your own history, your own fields, your own vocabulary, your own way of describing a decision. The resulting weights are yours: exportable, portable, and yours if you walk away. We would rather earn the renewal than hold your model hostage.
Prove it, do not trust it.
- Model cards. What it is, what it was trained on, what it is good at, and where it is weak.
- An eval harness. Switching or upgrading a model is a measurement, not a leap of faith.
- A full audit log. What the agent was asked, what it ran, what it returned.
- Governed access. The agent has read-only, scoped access. It cannot reach what it should not, and that is enforced at the data layer, not by asking it nicely.
What we will never do.
- We do not train on your data. If you ask us to, the model that comes out is yours.
- We do not meter your questions.
- We do not send your numbers to a third-party AI vendor.
- We do not sell your farm data to anyone, ever.
- We do not make price predictions.