Dedicated Instance
From subscription-tier dedicated to enterprise-grade customisation — one product line covering pro users to global enterprises
Hydite Vtslx AO Dedicated Instance ships in two clearly separated tiers:
🟦 Subscription Dedicated — single-tenant SaaS for power users and SMBs. Self-serve, monthly/annual subscription, zero ops.
🟪 Enterprise Dedicated — for large enterprises, government, finance, healthcare. Adds BYOK, self-hosted agent, private models, SSO, white-label, compliance and custom routing on top of the subscription tier. Available only on the Enterprise membership.
Both tiers expose the same OpenAI-compatible API, so migrating or upgrading is zero-code-change.
Tier comparison#
| Capability | 🟦 Subscription Dedicated (Personal / SMB) | 🟪 Enterprise Dedicated (Enterprise membership only) |
|---|---|---|
| Tenancy | Single-tenant | Single-tenant |
| Runs on | Hydite Cloud | Hydite Cloud / your cloud / your DC / sovereign stack |
| Models available | Hydite-curated pool | Curated + BYOK + private self-hosted |
| Routing | Dashboard presets | Full YAML / canary / header routing |
| Data residency | Hydite | Fully customer-controlled |
| Networking | Public + platform subdomain | Public / custom domain / VPC peering / Direct Connect / mTLS |
| Identity | Account + basic SSO | Enterprise SAML / OIDC / SCIM |
| Dashboard | Standard | White-label |
| Compliance | Standard ToS + DPA | HIPAA / SOC2 / MLPS-3 / sovereign / air-gap license |
| SLA | 99.9% | 99.95% baseline · 99.99% optional |
| Activation | One-click in dashboard | Sales + solution design + contract |
| Pricing | Monthly subscription, tiered by RPM/seats | Platform fee + usage (no usage cut on BYOK) |
| Time to live | < 5 min | 7–10 days (cloud) / 2–4 weeks (on-prem) |
Sections marked ✦ Subscription are baseline features; ✦ Enterprise sections require the Enterprise membership.
Part 1 · Subscription Dedicated ✦ Subscription
1.1 What it is#
Subscription Dedicated sits between Shared Edge and Enterprise:
- A single-tenant AO instance on Hydite multi-region cloud — isolated rate quotas, key namespace, cache and DB.
- vs Shared: no noisy-neighbour limits; QPS / TPM are allocated per tier and scale linearly.
- vs Enterprise: no choice of deployment location, no BYOK, no custom routing — everything runs on Hydite presets.
Best fit:
- Heavy solo developers / indie SaaS already saturating Shared Edge's 60 RPM.
- SMBs at 10M – 1B tokens / month that want stable first-token latency.
- Teams that need a branded subdomain like
acme.api.hydite.comwithout running infra. - Workloads with no hard data-residency requirements but wanting isolated quotas, billing and keys.
1.2 Subscription tiers#
| Tier | Audience | RPM cap | TPM cap | SSO |
|---|---|---|---|---|
| Pro (personal) | Senior devs / indies | 600 | 5M | Email / Google / GitHub |
| Team | 5–50-person teams | 3,000 | 30M | + Feishu / DingTalk / WeCom |
| Business | Mid-size businesses | 10,000+ | Negotiated | + basic Okta / Azure AD |
| Enterprise | Large / regulated | Unlimited | Unlimited | Full SAML / OIDC / SCIM |
Only the Enterprise tier unlocks the customisation features in Part 2 below. Pro / Team / Business subscribers — even when on Dedicated — do not have customisation rights and must upgrade to Enterprise.
1.3 Three-step activation#
Dashboard → Workspaces → Upgrade Plan:
- Pick a tier (Pro / Team / Business). The dashboard shows the resource sheet and monthly price.
- Pay (card / monthly invoice) and click Provision.
- ~5 min later your instance is live with a fresh subdomain (e.g.
acme.api.hydite.com). Point yourbase_urlat it.
Upgrades / downgrades are anytime — upgrade is immediate and pro-rated; downgrade takes effect at the next cycle (no quota refunds); any tier can be paused for up to 30 days.
1.4 Built-in features#
Every subscription tier ships with:
- ✅ Full API Reference
- ✅ Curated model pool (OpenAI · Anthropic · Google · DeepSeek · Qwen · Zhipu · Moonshot · xAI · Mistral …)
- ✅ Channels for env isolation (dev / staging / prod)
- ✅ Three-tier quotas (Key / Team / Org)
- ✅ Dashboard: Overview / API Keys / Channels / Billing / System Health
- ✅ Anomaly auto-pause for leaked keys
- ✅ Prometheus
/metricsand basic webhooks - ✅ 99.9% availability SLA
But not included: BYOK · self-hosting · custom routing · white-label · SAML/SCIM · advanced compliance. Those belong to Enterprise customisation.
Part 2 · Enterprise Dedicated ✦ Enterprise
🟪 Everything in this part is gated behind the Enterprise membership. Subscription tiers (Pro / Team / Business) — even on Dedicated — cannot enable any of the features below and need to upgrade to Enterprise first.
2.1 Why "customised"#
The Enterprise tier targets workloads where generic SaaS just doesn't fit:
- Compliance — finance, healthcare, gov, central / state-owned enterprises handling regulated or proprietary data.
- Compute sovereignty — must run private fine-tunes on your own GPU fleet.
- Commercial sovereignty — bill providers under your own contracts (BYOK), no Hydite usage cut.
- Brand sovereignty — white-label the AO console for resale to your own customers.
- Network isolation — only allow access via VPC peering / Direct Connect / private DNS.
- Operational boundary — upgrades, configs and security policies are owned by your security team.
2.2 Three deployment topologies#
| Topology | Control plane | Data plane (inference path) | Time to live | Best fit |
|---|---|---|---|---|
| A · Hydite Cloud Dedicated | Hydite cloud | Hydite cloud | 1–3 days | Want enterprise customisation but no self-hosting |
| B · Connected Self-Host (recommended) | Hydite cloud dashboard | Customer-hosted Agent | 5–10 days | Data must stay in VPC, but want Hydite's managed dashboard & upgrade pipeline |
| C · Air-Gapped Self-Host | Customer DC | Customer DC | 2–4 weeks | Fully offline, sovereign, classified, MLPS-3+ |
Topology B is our flagship pattern — keeps data inside the customer boundary while keeping ops on Hydite. Details follow.
2.3 Connected Self-Host: control plane / data plane split#
1┌───────────────────────────────┐ ┌──────────────────────────────────────┐2│ Hydite cloud control plane │ │ Customer VPC / DC │3│ control.hydite.com │ │ │4│ │ │ ┌──────────────────────────────┐ │5│ • White-label dashboard │◀───────┤ │ Hydite AO Agent (signed) │ │6│ • IAM / SSO / RBAC │ mTLS │ │ • outbound-only (no inbound)│ │7│ • Site mgmt + token issuing │ beat │ │ • pulls routing / blacklist │ │8│ • Routing config (YAML/Git) │ cfg │ │ • reports token / health │ │9│ • Usage rollup (metadata) │ push │ │ • local routing/cache/db │ │10│ • Audit / alerts / billing │ │ └────────────┬──────────────────┘ │11└───────────────────────────────┘ │ │ │12 │ ┌────────────▼──────────────────┐ │13 │ │ Your App │ │14 │ │ base_url=hydite-ao.svc:443 │ │15 │ └────────────┬──────────────────┘ │16 │ │ │17 │ ┌────────────▼──────────────────┐ │18 │ │ LLM providers (BYOK direct) │ │19 │ │ + private self-hosted models │ │20 │ └────────────────────────────────┘ │21 └──────────────────────────────────────┘Data flow matrix#
| Direction | Content | Through Hydite cloud? |
|---|---|---|
| App → Agent → LLM provider | prompt / completion / business data | ❌ never |
| Provider keys (BYOK) | encrypted in customer-local Vault | ❌ never |
| Agent → control plane | heartbeat, token counts, error codes, latency percentiles, virtual-key metadata | ✅ (mTLS, redactable) |
| Control plane → Agent | routing YAML, guardrail rules, key blacklists, version updates | ✅ (mTLS, push only) |
Business data stays 100% inside the customer boundary — the control plane sees only an "audit fingerprint" sufficient for billing, alerting and compliance audit.
Source-code & algorithm protection#
- The Agent ships as a distroless container image + Helm chart — no source code is shipped.
- Images come from
registry.hydite.com; pull secret is bound to the Site token and revoked on offboarding. - Cosign / Sigstore signed; signatures verified at boot against Hydite's public key — no anti-tamper bypass.
- Routing strategies, cache-key algorithms and pricing logic are pushed as WASM modules from the control plane — never persisted on disk locally.
Install flow#
Step 1 — issue a Site in the dashboard:
Dashboard → Sites → Create Site, fill name / compliance level / network mode. The system returns:
1SITE_ID=site_xxxxxxxxxxxxxxxxxxxx2SITE_TOKEN=hyt_site_eyJhbGciOi... # short-lived JWT, swapped for mTLS cert on first boot3HYDITE_REGISTRY_PULL_SECRET=...Step 2 — deploy the Agent in your K8s:
1helm repo add hydite https://charts.hydite.com2helm install hydite-ao hydite/ao-agent \3 --set site.id=$SITE_ID \4 --set site.token=$SITE_TOKEN \5 --set registry.pullSecret=$HYDITE_REGISTRY_PULL_SECRET \6 --set network.egress=via-corp-proxy # optional: route through corp egressStep 3 — outbound registration:
The Agent dials wss://control.hydite.com/agent, completes mutual TLS, the Site goes green in the dashboard.
Step 4 — cut traffic over:
1client = OpenAI(api_key=key, base_url="http://hydite-ao.acme.svc:443/v1")The full call path stays 100% on-prem / in-VPC with zero inbound ports — fully zero-trust compatible.
2.4 Air-gapped offline licence#
For fully isolated networks:
- Hydite issues a signed offline licence with expiry, RPM cap, model whitelist.
- The Agent verifies the signature + clock-drift tolerance at boot — no control-plane reachability needed.
- Config changes ride physical media: dashboard exports
config.yaml.signed→ copy to internal network → Agent loads it. - Usage ships back via offline reconciliation: Agent emits
usage-{date}.signed.json→ exported → uploaded to dashboard. - Sovereign-stack support: Kunpeng / Kylin / OceanBase / HuaweiCloud Stack / Zhongbiao Kylin / UOS.
2.5 BYOK and private model onboarding#
1# Typical routing config (Enterprise customisation only)2models:3 - alias: claude-sonnet-4-54 provider:5 name: anthropic6 model: claude-sonnet-4-57 api_key: ${ACME_ANTHROPIC_KEY} # your own Anthropic account8 api_base: https://api.anthropic.com9 tags: [prod]10 - alias: claude-sonnet-4-5 # same alias → automatic fallback11 provider:12 name: bedrock13 model: anthropic.claude-sonnet-4-514 region: us-east-115 tags: [fallback]16 - alias: acme-finetune-v3 # private self-hosted model17 provider:18 name: openai-compatible # any OpenAI-compatible endpoint works19 api_base: http://vllm.internal:8000/v120 tags: [private, on-prem]- Public-provider BYOK: bills under your contract — Hydite takes no usage cut, only the Enterprise platform fee.
- 100+ supported endpoints: Anthropic / OpenAI / Bedrock / Azure OpenAI / Vertex / DeepSeek / Qwen / Zhipu / Moonshot / xAI / Mistral / DashScope / Volcengine / Tencent / Baidu …
- Private model onboarding: vLLM / TGI / SGLang / LMDeploy / Triton / Ollama / sovereign stacks — anything OpenAI-compatible.
- Public and private models can share the same alias for canary / fallback (e.g. 70% private fine-tune / 30% Anthropic / fallback GPT-4o on timeout).
2.6 Custom routing strategies#
| Strategy | When to use |
|---|---|
| Latency-based | Multi-region, pick lowest P95 deployment |
| Cost-based | Cheapest equivalent model right now |
| Tag-based | Geo routing via tags: [cn, eu] |
| Header-based | Route by x-customer-tier: enterprise |
| A/B / canary | Gradual rollout 10% / 30% / 100% |
| Sticky session | Pin a user to one model to maximise cache hits |
All declarative YAML, visual editor in the dashboard, Git-versioned. Subscription tiers can only use presets.
2.7 Identity & access#
| Capability | Notes |
|---|---|
| Enterprise SSO | SAML 2.0 / OIDC (Okta · Azure AD · Feishu · DingTalk · WeCom) |
| SCIM provisioning | Auto-sync users, groups, deactivation |
| Granular RBAC | Org Admin / Team Lead / Developer / Viewer + custom |
| Audit trail | Every dashboard action is logged |
| API operation signing | HMAC + 2-step confirm for destructive ops |
| Network ACL | Restrict dashboard / API to source IPs / CIDRs / VPN |
Subscription tiers only get basic SSO (see 1.2). SAML / OIDC / SCIM is Enterprise-only.
2.8 Networking#
Mix and match any of:
- Public + custom apex —
https://ai.acme.com/v1with auto-issued or BYO certs. - VPC Peering / PrivateLink — drop an ENI / endpoint inside your VPC, traffic never traverses the public internet.
- Direct Connect — Aliyun Express Connect / AWS Direct Connect / Azure ExpressRoute.
- IP allow-list + mTLS + zero-trust egress (Zscaler / Cloudflare Access).
- Multi-region dedicated — e.g. primary in us-east, DR in eu-west, branch in ap-southeast — cross-region fallback handled by AO routing.
Subscription tiers only support "public + platform subdomain".
2.9 Capacity, performance & SLA#
| Metric | Subscription | Enterprise |
|---|---|---|
| RPM / TPM | Tiered cap | Unlimited (per cluster capacity) |
| Sustained QPS / replica | < 1k | 5k+, linear scale-out |
| P50 first-token | < 50ms | < 30ms |
| Availability SLA | 99.9% | 99.95% baseline · 99.99% optional |
| RTO / RPO | 30 min / 1 h | 5 min / 0 (multi-AZ) · 30 s / 0 (multi-region) |
| Outage credits | — | Pro-rated against monthly fee |
2.10 Observability#
Both Subscription and Enterprise include the five dashboard surfaces (Overview / Anomaly / API Keys / Channels / System Health) and /metrics. Enterprise adds:
- White-label dashboard — your logo, theme and domain (e.g.
console.acme.com); resellable. - Structured audit logs — every call + dashboard event streamed as NDJSON to your S3 / OSS / Splunk / Elasticsearch / Aliyun SLS.
- Replay & debug — once Logging is enabled, inspect any past call's prompt, completion, token counts and upstream timings.
- BYO KMS — at-rest encryption uses your CMK (AWS KMS / Azure Key Vault / GCP KMS / HashiCorp Vault).
2.11 Onboarding flow#
A typical 4-step rollout — first business traffic in 7–10 working days:
- Discovery (D+0–3) — solution architect aligns on topology, compliance, capacity, network, IDP / SIEM / provider list.
- Environment delivery (D+3–7) — Hydite cloud same-day; customer-cloud via IaC modules; on-prem via offline package. SSO, domain, certs, KMS, private models all wired.
- Integration & load test (D+7–10) — compatibility suite + replay-based load test, validation of RPM / latency / cost targets.
- Cutover — flip
base_url, canary then full rollout. Zero client-code changes.
Ongoing operations are owned by Hydite Customer Success: quarterly capacity review + bi-annual DR drill + provider change advisory + dedicated Slack / Feishu channel + 24×7 P1 SLA.
2.12 Commercial model#
Dual-track: monthly platform fee + usage.
- Platform fee covers infra, dashboard, SLA and customer success — tiered by cluster size and SLA target.
- Usage is billed on actual token consumption. Hydite-procured providers ride our volume discount; under BYOK Hydite charges only a thin routing/audit fee.
- Professional services (optional): fine-tuning, agent workflow co-build, on-site compliance audits, sovereign-stack porting.
Pricing always starts with a 30-minute solution call. Trigger via Workspaces → Upgrade to Enterprise in the dashboard, or email enterprise@hydite.com.
Upgrade path
1Shared Edge (free / pay-go) → Sub. Pro → Team → Business → Enterprise Dedicated2 ↑ ↑ ↑3 single-tenant isolation basic enterprise SSO BYOK / self-host / white-label / SAMLEvery step is zero-code-change:
- Shared → Subscription Dedicated — one click in the dashboard; keys, channels and usage history migrate.
- Subscription → Enterprise Dedicated — once the contract is signed we either migrate the cluster to Enterprise or stand up a Connected Self-Host alongside it; both run dual-active during cutover, old keys move via
/key/migrate.
Next steps#
- Browse every endpoint → API Reference
- Try Subscription Dedicated → Workspaces → Upgrade Plan in the dashboard
- Talk to Enterprise sales → enterprise@hydite.com
- Compare with Shared → Shared Edge Instance