The AI-first cloud · v0.4 agent runtime is live

Infrastructure your
agents can operate.

LiveLLM gives you real Kubernetes, VMs, browsers and Postgres — then hands each one to AI agents as a tool they can drive. Every move an agent makes lands as a Git commit you can review, diff and roll back, like code.

AI-firstEvery resource is a toolVersioned by Git
Send your AI agent to LiveLLMcloud.live-llm.com/SKILL.md
cloud.live-llm.com/acmeReady
Resources
4 running
webcontainer app
web-8080-acme.live-llm.com
Open
🖥
opslinux vm
ssh ops-22-acme.live-llm.com
SSH
🔎
researchbrowser
research-vnc-acme.live-llm.com
Open
🗄
mainpostgres
main-db-acme · internal
Connect
…or call the API
$ curl api.live-llm.com/v1/tenants/acme/deploy \
   -d '{"app":"web"}'
✓ web-8080-acme.live-llm.com
Activity · git history
agent·reviewer commented on PR #46
2m
agent·builder deployed web · sha-9fk2
5m
you merged #45 · add db replica
1h
agent·research opened PR #44 · invoices
3h
What we stand for

Three things, on purpose.

AI-first

Agents are first-class workloads — their own runtime, your workspace identity, and the same API as everything else.

A master agent orchestrates a whole fleet
Pluggable models — Claude, OpenAI or self-hosted
Build an entire app by describing it

Every resource is a tool

VMs, browsers, containers and databases expose control surfaces an agent can call. No glue code, no key-juggling.

An agent drives a VM over SSH
Hand a browser to an agent as a tool
Agents inherit the workspace's secrets & access

Versioned by Git

A Git org per workspace, continuously reconciled. Every action — human or agent — is a commit or PR.

Review, diff and roll back like code
Build from source on every push
Audit every change, human or agent
See it work

Agents with hands, not just chat.

Pick a job. Watch an agent drive a real resource and leave a reviewable trail.

browser · review mode

Drives a real browser, step by step.

Hand a browser workload a plain-language task. The agent plans it, acts, and streams back a trajectory you can replay — or pause for a verdict at each step.

Review mode pauses for a human verdict before each action.
Full session recording, plus per-step screenshots.
Runs scoped to the workspace — no shared credentials.
task_9fk2 · trajectory
› plan: "Download June invoices"

  ✓ 0  open billing portal
  ✓ 1  filter to June 2026
  ⏸ 2  download 3 PDFs   ← awaiting verdict
      [approve]  [reform]  [skip]

  result: pending review
The platform

Real primitives. Nothing you can't reach.

Every resource is an actual cluster object you can kubectl into, SSH to, and watch — wired together so a workspace is a working stack, not a shopping list.

Agent runtime

Long-running, scoped agents. A master delegates to a fleet through their own API.

mastersfleetsautonomous
🎞

Reviewable browser agents

Every run is a replayable trajectory with per-step verdicts and a recording.

trajectoryreviewrecording

Real Kubernetes

An isolated cluster per workspace, with its own API server.

🖥

VMs & desktops

Ubuntu, Windows and GPU desktops — reach them over SSH, RDP or VNC.

🗄

Managed data

HA PostgreSQL, S3-compatible object storage and scheduled snapshots.

📡

Deep observability

Traffic, latency, logs and live metrics — scoped to your workspace.

🔐

Isolated by default

Network isolation, managed secrets, policy guardrails and signed images.

🔗

Reach anything

Per-port HTTPS on one wildcard domain, gated by IP allowlist or auth.

Build from a sentence

Describe an app; it scaffolds, builds from source and deploys live.

from gitautodeploy
🔔

Alerts that reach you

Webhook, Telegram or email — per workload, on your thresholds.

Integrations & features

Everything your product plugs into.

ClaudeOpenAISelf-hosted modelsPostgreSQLObject storageKubernetesGPULinux VMsWindows VMsContainersManaged GitBuild from sourceSSHRDPVNCBrowser agentsBrowser APIWebhooksTelegramEmailClaudeOpenAISelf-hosted modelsPostgreSQLObject storageKubernetesGPULinux VMsWindows VMsContainersManaged GitBuild from sourceSSHRDPVNCBrowser agentsBrowser APIWebhooksTelegramEmail
Models3
Claude
Anthropic frontier models
OpenAI
GPT, embeddings, realtime
Your own models
Self-hosted on your GPUs
Data3
PostgreSQL
HA databases in one click
Object storage
S3-compatible buckets
Backups
Scheduled snapshots
Compute3
Kubernetes
A cluster per workspace
Linux & Windows VMs
Full machines, GPU optional
Containers
Any image, instant URL
Dev & Git3
Managed Git
A repo org per workspace
Build from source
Push to deploy
Templates
Save & reuse setups
Automation3
Browser agents
Reviewable web automation
Automation API
Scrape, fill, extract
Schedules
Cron-driven agent jobs
Access3
SSH · RDP · VNC
Reach any machine
Per-port HTTPS
Every port gets a URL
Access control
IP allowlists & auth
Notifications3
Webhooks
POST on every event
Telegram
Alerts to your chat
Email
Digest & incident mail

Give AI a full dev
environment. Let it cook.

Spin up a workspace, point an agent at it, and let it build, deploy and ship — every step in Git, every resource yours to take over.