v0.4 — Agent runtime preview is live

The classical cloud,
agent-native.

Kubernetes clusters, VMs, Postgres, object storage and a fleet of AI agents — all provisioned from a single spec, all delivered by GitOps. The infrastructure your engineers expect, on top of a runtime your agents understand.

One spec file. One pull request. Everything else is GitOps.

tenants/acme/tenant.yaml

apiVersion: livellm.cloud/v1
kind: Tenant
metadata:
  name: acme
spec:
  plan: team
  workload:
    type: kubernetes     # or vm-ubuntu | vm-windows
    ha: true
  integrations:
    - postgres           # CNPG cluster
    - object-storage     # MinIO bucket
    - gitea-org          # free SCM org
  agents:
    - role: code-reviewer
      model: claude-opus-4-7
    - role: data-ingestor
      schedule: "@hourly"

{ }

Classical when it should be

Real namespaces, real pods, real network policies, real DNS. Nothing magical you can't kubectl into.

Per-tenant vcluster with its own API
Wildcard cert, isolation by NetPol
Postgres, MinIO, Longhorn — operator-grade
ArgoCD App-of-Apps, fully observable

⌬

Integrated by default

Twenty-something add-ons wired together so a tenant gets a working stack, not a shopping list.

Vault + ExternalSecrets, no manual mounts
Cert-manager + DNS-01, no renewal Mondays
Per-tenant Gitea org with bootstrap repo
Stripe / YooKassa, RU-aware billing

Built for agents

The future has long-running, scoped, billable AI agents. We are designing the runtime today.

Agents inherit the tenant's identity
Models pluggable: Claude · OpenAI · vLLM
Every action becomes a Git pull request
Same kubectl-style spec your humans use

Integrations

Twenty-something cloud primitives.
One spec to rule them.

Everything below is wired up in this cluster today. You don't install operators, write Helm values, or stitch CRDs — you toggle a field, open a PR, and ArgoCD rolls it in.

Compute3

Kubernetes
Isolated vclusters per tenant
KubeVirt
Ubuntu & Windows VMs side-by-side with pods
Knative
Scale-to-zero functions (coming soon)

Data3

CloudNativePG
HA Postgres clusters in one line of spec
MinIO
S3-compatible object storage
Longhorn
Block storage with snapshots & replicas

Networking3

Traefik
Wildcard *.cloud.live-llm.com ingress
cert-manager
Automatic Let's Encrypt DNS-01
NetworkPolicy
Tenant isolation by default

Identity3

Vault
Per-tenant secret stores via ExternalSecrets
Kyverno
Policy-as-code, signed images, baseline PSS
Authentik
OIDC SSO (rollout in Phase 3.5)

Delivery3

ArgoCD
Every tenant is a GitOps Application
Gitea
Free SCM org per tenant
GitHub
Bring your own Git (Phase 7)

Observability4

Prometheus
Per-tenant metrics scoped to namespace
Grafana
Pre-baked dashboards on day one
Loki
Structured logs, label-isolated
OpenCost
Per-namespace dollar attribution

AI3

Claude
Anthropic's frontier models, billed through tenant
OpenAI
GPT, embeddings & realtime APIs
vLLM
Self-hosted OSS models on tenant GPUs

Billing2

Stripe
Cards & invoices outside RU
YooKassa
RU-domestic payments

The vision

Every tenant ships with a team of agents.

Classical clouds give you CPU, memory and disk. We give you all of that and a persistent runtime where AI agents review code, ingest data, watch dashboards, open pull requests and respond to incidents — scoped to your tenant, billed by your tenant, never crossing the border.

Code reviewer

Hooks into every PR opened in the tenant Gitea org. Comments, suggests fixes, opens follow-up PRs. Models pinned per plan.

agent: code-reviewer model: claude-opus-4-7 scope: pull_requests

Data ingestor

Cron-driven jobs that pull from sources you describe in YAML and land into the tenant's Postgres or object bucket. Schema-aware.

agent: data-ingestor schedule: "@hourly" sink: postgres://main

Incident responder

Watches Grafana alerts + Loki logs. Triages, links runbooks, opens an issue. If you authorise it, rolls back a Deployment via PR.

agent: incident-responder alerts: grafana://* authority: propose

Schema migrator

Reads diffs against your ORM, generates safe migrations, proposes them through Gitea, executes after approval.

agent: schema-migrator source: prisma|alembic approvals: 1

Cost watchdog

Reads OpenCost hourly, flags drift over your budget, opens a PR that scales the offending Deployment or shifts to spot nodes.

agent: cost-watchdog budget: $50/day authority: propose

Customer copilot

Sits inside your product as an SDK. Inherits the tenant's policies and secrets — no key-juggling, no separate billing.

agent: customer-copilot sdk: @livellm/agent billing: inherits

Roadmap

Phase 0Foundationsshipped
Vault · ESO · Kyverno · cert-manager · KubeVirt · CNPG · MinIO · Longhorn · observability
Phase 1Tenant CRshipped
namespace + quota + NetPol + cert + AppProject + vcluster (HA-aware)
Phase 2tenant-api PR machineshipped
Go service that opens GitOps PRs from a spec
Phase 3tenant-ui consoleshipped
this site — wizard, dashboard, billing surface
Phase 4Billing & meteringnext
OpenCost → plans table → Stripe/YooKassa invoices
Phase 5Agent runtimenext
First-class CRDs for agents, tenant-scoped secrets & SDK
Phase 6VM workloadsnext
Ubuntu & Windows side-by-side with K8s tenants
Phase 7External Gitlater
Bring your own GitHub/GitLab/Gitea
Phase 8Hardeninglater
Velero, pgBackRest, rate limits, status page, audit

Open a pull request.
Get a cloud.

Live LLM Cloud is in early access. Founders get the platform for free until billing ships — then locked-in plan, no surprise pricing.

Get founder access Talk to the founder

The classical cloud,agent-native.