Enterprise deployment

It runs on your side of the network.

Snyte ships as containers with the deployment manifests committed to the repository. There is no Snyte-operated cloud to review, because every path is one you run. That is a real constraint as much as a feature, and the models below say what each one asks of your team.

Deployment models

Three paths, all self-managed.

Pick one to see what it brings up and what it needs from you.

The full production path. Horizontal autoscaling, network policies, and external secret references are committed as manifests rather than described in a runbook.

Committed artifacts: k8s/ — 20 manifests including hpa.yaml, network-policy.yaml, ingress.yaml
You provide: A cluster, an ingress controller, a secrets backend for the ExternalSecret path, and TLS material you manage.
Know this first: Snyte does not operate this for you. There is no Snyte-hosted cloud and no BYOC product; the cluster and its etcd encryption are your responsibility.

Application
Data store
Operations

webNext.js standalone server, non-root
apiGo service, distroless image
workersAnalytics and forecasting jobs
postgresPrimary store; RLS requires a non-superuser role
clickhouseTime-series analytics
redisCache and rate-limit state
hpaHorizontal pod autoscaling
network-policyPod-to-pod traffic restrictions
external-secretsSecrets pulled from your backend
monitoringPrometheus and Grafana manifests

Governance

Controls your platform team will ask about.

Single sign-on

WorkOS AuthKit with JWKS-based RS256 validation, plus TOTP two-factor. Directory sync via SAML or SCIM is not available yet.

workos_auth.go

Roles and permissions

Role hierarchy enforced per route (admin, editor, viewer) with per-action permission checks and ownership verification on data access.

authorization middleware

Audit trail

Access, authentication, and security events written to one log, queryable per user, per organization, and system-wide.

audit_service.go

In-perimeter model serving

Point the platform at a self-hosted Ollama model and no prompt, schema, or result row leaves your network. Hosted providers remain an option, not a requirement.

ai/providers/ollama.go

The full control matrix, including what is not in place

Rollout

How an evaluation actually goes.

Four steps, in this order. The first one is the one that matters: if the trace does not convince your reviewers, nothing after it will.

Technical session

One of your sources, one of your questions, your reviewers watching the trace. This is the step that decides the evaluation.

Security review

Start from the published posture and the gaps it lists. Full assessment report and incident response plan available under NDA.

Pilot inside your perimeter

Compose on a single host or a namespace in your cluster, connected to a real source with a bounded user group.

Production rollout

Kubernetes manifests, autoscaling, network policy, and your secrets backend. Migrations run as a separate step by default.

Commercials

No price list, on purpose.

Self-serve billing has not shipped, so a published tier table would be a number we cannot yet honour. Pricing is scoped per deployment: the number of connected sources, the size of the user group, and whether you run hosted or in-perimeter model serving.

Talk to sales How it works

What we will not do

Quote an uptime SLA we have no document for.
Promise a dashboard migration the import pipeline cannot perform.
Show you a customer logo wall. There is nothing to show yet.