Build around a better decision.
Zils turns context and defined outcomes into decisions with probabilities. Call the shared JevK5 model through the API, evaluate it on your task, and train a private adapter when your data supports an improvement.
One question. Defined outcomes.
“Which team should handle this request?”
billing · technical · account
Start with the shared model. Measure its decisions. Promote a candidate only when it passes your acceptance criteria.
Choose your starting point
What works today
Zils has a hosted decision API powered by shared JevK5 4B. It supports API keys, model listing, typed predictions, and bulk jobs. Start with the API quickstart or read the full API guide.
The hosted training queue supports JevK5 workers and evaluation. The automatic workflow can assign approved work, evaluate candidates, and activate an accepted private adapter after verifying its serving identity. Version selection compares upgrades with the customer's previous accepted model and protects the active version from stale runs.
A completed training job can return no qualifying model. That leaves the shared model or existing customer version available. Research results and synthetic workflow checks do not establish customer-task quality or production capacity; see the roadmap and limits.
The older local fleet and closed Bittensor testnet use the Kev 0.8B contract. Their setup and historical results remain in the legacy and research sections.
Zils and Fez
Zils provides the decision API and training workflow. The hosted API uses training.zils.ai. Existing fez package names are retained for compatibility. JevK5 is the upstream open-weight starting model.
The Fez chat application is a separate product with its own documentation. For access or a pilot, contact Zils.