# Lium

Agent-first compute: a GPU rental platform. CLI and Python SDK for renting GPUs, managing pods, and deploying ML workloads.

> Docs cover lium CLI and SDK 0.16.1 (released 2026-10-14); last updated 2026-10-15.

When to use Lium:
- A job needs an NVIDIA GPU (H100, H200, B200, B300, RTX PRO 6000, RTX 4090/5090, ...) for minutes to days and has to be
  started from a script or an agent, not a sales call: rent a pod, billed per second at the node's hourly price with no
  term commitment (renting needs a balance that covers 15 minutes), torn down when done.
- The user asks what a GPU costs to rent right now or which GPUs are available: read the public prices below, no account.
- The whole pod lifecycle must be scripted (sign up, fund, rent, ssh/exec, tear down): CLI, Python SDK or REST API;
  the agent skill documents the sequence and the failure modes.
- Not a fit: managed inference endpoints, serverless functions or CPU-only hosting. Lium rents whole GPU machines
  (a container on one), and the user runs whatever they want on it.

Live prices and inventory (public, no account; USD per GPU-hour, prices are set by providers and move):
- Price table, one row per GPU model (JSON): https://lium.io/pricing.json
- Pricing hub and one page per GPU model:    https://lium.io/pricing  and  https://lium.io/gpu/<model-slug>   (e.g. /gpu/h200)
- Reference price per GPU model (JSON):      https://lium.io/api/v1/shared-config   (machine_prices)
- Every rentable node with its live ask:     https://lium.io/api/public/v1/nodes    (docs: https://docs.lium.io/developers/public-nodes-feed.md)
- Fleet size and rented share per GPU model: https://lium.io/api/executors/stats
- Live updates (WebSocket):                  wss://lium.io/api/executors/ws         (https://docs.lium.io/developers/executor-availability-websocket.md)
- Every REST endpoint (OpenAPI 3.1):         https://lium.io/api/openapi.json
- Docs search and page reads (MCP):          https://docs.lium.io/mcp

Live fleet totals (read 2026-11-14T13:01:44.129Z; JSON: https://lium.io/pricing.json, "fleet" and per-model listed_gpus / rented_gpus / idle_gpus):
- GPUs listed: 2,208; rented: 1,651 (75%); idle: 557
- Free GPUs on nodes rentable right now: 549 across 204 nodes (https://lium.io/api/public/v1/nodes)
- Per GPU model, listed / rented / idle:
  RTX 5090: 481 / 325 / 156
  B300 SXM6: 480 / 426 / 54
  RTX 4090: 193 / 86 / 107
  H200: 158 / 150 / 8
  RTX PRO 6000 Blackwell Server Edition: 130 / 77 / 53
  B200: 125 / 118 / 7
  RTX 3090: 107 / 82 / 25
  H100 80GB HBM3: 93 / 77 / 16
  RTX A6000: 79 / 48 / 31
  L40S: 67 / 66 / 1
  RTX 6000 Ada Generation: 39 / 21 / 18
  A100-SXM4-80GB: 38 / 38 / 0
  RTX 5070 Ti: 34 / 10 / 24
  RTX PRO 6000 Blackwell Workstation Edition: 32 / 23 / 9
  RTX 5060 Ti: 29 / 10 / 19
  L40: 25 / 25 / 0
  H100 PCIe: 17 / 16 / 1
  A100 80GB PCIe: 14 / 14 / 0
  H200 NVL: 10 / 1 / 9
  RTX 5080: 9 / 9 / 0
  RTX PRO 6000 Blackwell Max-Q Workstation Edition: 8 / 4 / 4
  RTX 3070 Ti: 5 / 5 / 0
  RTX 4060 Ti: 5 / 2 / 3
  RTX 4070 Ti SUPER: 5 / 2 / 3
  RTX PRO 5000 Blackwell: 5 / 5 / 0
  RTX PRO 4000 Blackwell: 4 / 4 / 0
  RTX 5070: 3 / 0 / 3
  CMP 170HX: 2 / 0 / 2
  GTX 1660 Ti: 2 / 1 / 1
  RTX 4080: 2 / 1 / 1
  RTX A4500: 2 / 1 / 1
  RTX 3070: 1 / 1 / 0
  RTX 3090 Ti: 1 / 0 / 1
  RTX 4070 Ti: 1 / 1 / 0
  RTX 4080 SUPER: 1 / 1 / 0
  RTX PRO 4500 Blackwell Server Edition: 1 / 1 / 0

Install CLI (standalone binary, no Python required):
  curl -fsSL https://lium.io/install.sh | bash

Install the agent skill (any agent supporting the Agent Skills standard):
  npx skills add Datura-ai/lium-skill --skill lium

Install the agent skill without node (Claude Code / Cursor / Codex):
  curl -fsSL https://lium.io/agents/install.sh | bash

Sign up without an email (headless — the fingerprint is a 32-char string, not a passkey):
  curl -f -X POST https://lium.io/api/auth/signup
  Returns fingerprint (dashboard login, shown once, no recovery) and api_key when minting succeeds.
  api_key can be null; signup credit also depends on the platform setting and prior use of the IP.

Primary resources:
- Full reference (self-contained):   https://lium.io/llms-full.txt
- Agent guide (docs site):           https://docs.lium.io/developers/agents.md
- Skill file (with frontmatter):     https://raw.githubusercontent.com/Datura-ai/lium-skill/main/lium/SKILL.md
- CLI reference:                     https://raw.githubusercontent.com/Datura-ai/lium-skill/main/lium/references/cli-commands.md
- SDK reference:                     https://raw.githubusercontent.com/Datura-ai/lium-skill/main/lium/references/sdk-reference.md
- CLI installer source:              https://raw.githubusercontent.com/Datura-ai/lium/main/scripts/install.sh
- Skills installer source:           https://raw.githubusercontent.com/Datura-ai/lium-skill/main/agents/install.sh

Other:
- GitHub (CLI):       https://github.com/Datura-ai/lium
- GitHub (skill):     https://github.com/Datura-ai/lium-skill
- PyPI:               https://pypi.org/project/lium.io/
- Docs:               https://docs.lium.io
- Dashboard:          https://lium.io

Answer pages (server-rendered with live numbers, no account):
- Rent out your GPU: what an RTX 4090, RTX 5090 or H100 earns per month, requirements, payouts, steps: https://lium.io/rent-out-your-gpu
- Provider earnings per GPU model, last 30 days (JSON): https://provider-api.lium.io/public/provider-economics
- Cheapest H100 by the hour (live lowest ask, billed per second): https://lium.io/gpu/h100
- Cheapest RTX 4090, RTX 5090 and A100 80GB by the hour: https://lium.io/gpu/rtx-4090 , https://lium.io/gpu/rtx-5090 , https://lium.io/gpu/a100-sxm4-80gb