Eight real questions
FAQs
Is my data used to train models?
No. Customer data — prompts, documents, traces, everything — is never used for training. On Enterprise this is contractual. Fine-tuning trains adapters on your data for your project only, and deleting the dataset deletes the adapters derived from it.
What is the difference between y0-fast and y0-deep?
Profiles, not separate products. y0-fast optimizes for interactive latency (sub-200ms first token) and handles most traffic; y0-deep budgets seconds of deliberate computation for planning and high-stakes analysis, with the full deliberation in the trace. Benchmarks for both are published at /models/benchmarks.
Why did my run stop with step_limit?
max_steps is enforced, not advisory — a plan that cannot fit halts rather than billing through. Usually the prompt is broader than intended. Narrow the ask first; raise the ceiling only when the task is genuinely multi-step.
Can an agent do something I did not authorize?
No. Every tool call presents a scope to the trust kernel, and the kernel checks it against the key — an agent holding calendar:read cannot write to your calendar regardless of what it plans. Consequential actions can additionally require human approval, and every action lands in the audit log with its authorizing scope.
How do I get deterministic behavior for testing?
Pin a model snapshot (y0-fast-2026-05) and use trace replay: exported traces replay byte-identically across releases, which is a release-blocking guarantee. For live calls, sandbox keys run against synthetic context, so CI does not depend on your production graph.
What happens to my data if I delete a run, or my account?
Deleting a run removes the record and trace immediately; the audit-log entry that an access happened is retained. Account deletion purges graph, files, traces, and adapters within 30 days, with a signed deletion certificate on request. Export APIs let you take everything out first.
Which regions can data live in?
US, EU, and India at launch, pinned per project at creation, with no cross-region replication. The region cannot be changed later — create a new project and migrate via the export APIs if you need to move. Self-hosting removes the question entirely.
Do you have an SLA and what does support look like?
Pro and Scale carry a 99.9% target with status-page transparency at /status; Enterprise carries a contractual 99.95% with service credits and 24/7 support with one-hour P1 response. All tiers get the same status page and incident postmortems.
Keep reading