ZeroGrad
Capabilities · Recommendations

Product Recommendations

Healthy
In-context predictionEndpoint product-recommendationsProduction v12Challenger v13

Production constraints

Constraints drive candidate selection and deployment decisions.

P95 latency
<
ms
Availability
>
%
Catalog coverage
>
%
Cost / 1K
<
$
Challenger v13: satisfies all constraints
Candidate eligibility preview
✓Managed ANDCG@10 .412 · 58ms · $0.22Selected
✕Managed BNDCG@10 .426 · 112ms · $0.61P95 112ms, cost $0.61
✓Managed CNDCG@10 .338 · 11ms · $0.03Eligible

Capability

Endpoint
product-recommendations
Task
Implicit-feedback ranking
Objective
Recommend products each user i…
Mode
In-context prediction
Backend
Managed by Zero Grad
Examples
interactions.parquet
Available history
46.2M observations
In-context examples
50 / request
Target
purchased
Production version
v12

By default you manage capabilities and versions, not models. Turn this on to see which engine family backs each candidate.

Optimization mode

Preview
Optimize for
Allowed actions

When a better model appears, Autopilot evaluates it against your data, shadows it, and rolls it out within your constraints — your application changes zero code.

Where should this capability run?

Enterprise controls
Data residency
Pin data and inference to a region (US, EU)
Enterprise
Private networking
PrivateLink / Private Service Connect endpoints
Enterprise
Customer-managed encryption keys
Bring your own KMS keys
Coming soon
Audit logs
Every deployment, promotion and config change, exportable
Enterprise
SSO / SAML
Okta, Google Workspace, Azure AD
Enterprise
RBAC
Separate who can deploy, promote and change guardrails
Enterprise
Dedicated endpoints
Isolated capacity with reserved throughput
Enterprise
Custom retention
Configure how long inputs and predictions are kept
Coming soon

Compliance documentation is shared directly with design partners on request.