Your compliance, your models, your cluster.
Everything in Growth, plus the controls required to deploy distributed AI inside regulated, air-gapped, or contractually sensitive environments.
Bring-your-own-model (BYOM)
Upload a GGUF or ONNX model (≤14B parameters) to your private registry. MeshInfer.AI handles quantization into Q4/Q5/Q8 variants, signs each bundle with your Enterprise key, and distributes to only your tenant's nodes.
| Supported architectures | Llama 2/3/3.1/3.2/3.3 · Qwen 2/2.5 · Mistral · Phi-3/3.5 · Gemma 2 · any GGUF-compatible |
| Upload via | Dashboard or `meshinfer model push ./my-model.gguf --tenant acme` |
| Quantization | Managed pipeline — GGUF Q4_K_M / Q5_K_M / Q8_0 produced automatically |
| Distribution | Signed bundles; only devices bearing your tenant certificate can decrypt |
| Eval harness | Built-in MMLU/IFEval/your-private-evals comparison against cloud baselines |
On-prem coordinator
Deploy the entire control plane inside your VPC or data center. MeshInfer.AI SaaS becomes optional — ideal for air-gapped / FedRAMP High workloads.
helm repo add meshinfer https://charts.meshinfer.ai
helm install meshinfer meshinfer/coordinator \
--namespace meshinfer --create-namespace \
--set tenant=acme \
--set license=$LICENSE_KEY \
--set storage.class=premium-rwo \
--set auth.mode=saml \
--set auth.saml.metadataUrl=https://sso.acme.com/saml/metadata| Footprint (baseline) | 3 nodes · 8 vCPU · 16GB RAM · 200GB SSD each |
| Dependencies | Postgres 15+ · Redis 7+ · NATS JetStream 2.10+ · object store (S3/GCS/MinIO) |
| Updates | Quarterly GA, monthly patch, signed and attested |
| Phone-home | Off by default. Metrics exported to your Prometheus only. |
SSO & identity
| Protocols | SAML 2.0 · OIDC · OAuth 2.0 |
| IdPs tested | Okta · Azure AD · Google Workspace · OneLogin · Ping · JumpCloud |
| SCIM 2.0 | Automated user/group provisioning & deprovisioning |
| RBAC | Built-in roles: Owner · Admin · Developer · Auditor · Billing. Custom roles via policy DSL. |
| Per-project API keys | Scoped, rotatable, with IP allow-lists and model allow-lists |
Audit logs
Every control-plane action and every inference metadata event is written to an append-only, hash-chained audit log. Exportable to SIEM in near-real-time.
{
"ts": "2026-04-20T14:22:03Z",
"actor": { "type":"user", "id":"u_12", "email":"eng@acme.com" },
"action": "policy.update",
"target": { "type":"policy", "id":"pol_7" },
"diff": { "before": {...}, "after": {...} },
"ip": "10.0.4.18",
"request_id": "req_01HX...",
"prev_hash": "blake3:...",
"hash": "blake3:..."
}Exporters ship to Splunk, Datadog, Elastic, and generic syslog out of the box.
Custom SLAs
| Availability | Up to 99.99% (standard 99.95%) |
| p95 routing latency | Committed ≤ 15ms at gateway |
| p95 cloud-fallback latency | Provider-adjusted; published monthly |
| Support | 24/7 on-call engineer, 15-minute response on P0 |
| Credits | Pro-rated credits for any SLO miss, auto-applied next invoice |
Dedicated routing clusters
Single-tenant Coordinator deployment in your preferred region(s) — including EU-only, UK-only, or APAC-only routing for data-residency requirements. No shared compute, no noisy neighbors.