NEWGPU owners: host free demos on your card and earn a bigger share of every paid job. See how →

For AI agents

A private LLM API for your AI agent

Point OpenClaw, Hermes Agent or any OpenAI-compatible agent at open models with tool calling and long context. Prompts and outputs are never stored, and one switch keeps every request on machines we own.

Get an API keyPaste this into Claude Code, Cursor or any coding agentIt reads the agent quickstart, asks you for your key, and makes one test call.OpenClaw setupHermes Agent setup
The short answer. Base URL https://api.legit2compute.com/nodes/llm/v1, your key, and ornith-1.5:9b, a model we keep loaded at all times. Switch on Private in your dashboard and every request is served only by machines we own and operate, at 1.5× the listed price. Try it right now with the demo key below, no account needed.

What Private means

Nothing stored

We record the model, token counts, latency and status. Never the prompt, never the output, on either tier.

Our machines only

With Private on, requests are served only by machines we own and operate. Never a community provider's node, never an outside service, no fallback to either.

Visible on every reply

Each response carries X-L2C-Privacy: private or shared, and your activity page lists every request with its tokens and cost.

Models that suit agents

Native tool calling and 64K+ context, read from the live catalog. Dollars per million tokens, input / output.

ModelContextSharedPrivateLoaded
Ornith 1.5 9B (vision, 256K)
ornith-1.5:9b · image in
256K$0.080 / $0.14$0.12 / $0.21 ● always on
Llama 3.1 8B
llama3.1
128K$0.030 / $0.050$0.045 / $0.075 ● always on
Ornith 1.5 35B (vision, 256K)
ornith-1.5:35b · image in
256K$0.12 / $0.80$0.18 / $1.20 ● loaded
Bonsai 2 9B (vision)
bonsai-2:9b · image in
256K$0.064 / $0.10$0.096 / $0.16 ○ on demand
MiMo-V2.6 9B (vision)
mimo-v2.6:9b · image in
256K$0.030 / $0.10$0.045 / $0.15 ○ on demand
LFM3 1.7B (text)
lfm3:1.7b
256K$0.040 / $0.060$0.060 / $0.090 ○ on demand
Gemma 4 12B (vision)
gemma4:12b · image in
128K$0.040 / $0.12$0.060 / $0.18 ○ on demand
Qwen3.8 27B (vision + text)
qwen3.8:27b · image in
128K$0.15 / $1.80$0.22 / $2.70 ○ on demand

Always on models are kept loaded, so your agent's first call answers straight away. Every model →

Try it in thirty seconds

curl https://api.legit2compute.com/nodes/llm/v1/chat/completions \
  -H "Authorization: Bearer l2ck_demo_cc7af91fbc644562a91444be9c80ce81" -H "content-type: application/json" \
  -d '{"model":"ornith-1.5:9b","messages":[{"role":"user","content":"In one sentence: why run an agent on a private model?"}]}'

That is our public demo key: free, shared by everyone, rate-limited, and it calls ornith-1.5:9b only, on the shared tier. Your own key calls every model, has its own limits and can switch on Private; signing up with Discord, Google, GitHub or Twitch includes $1 of trial credit.

Set up your agent

OpenClaw

Add us as a provider in openclaw.json, then talk to your assistant from Discord, Slack, Telegram or WhatsApp.

Hermes Agent

One wizard step makes us its custom endpoint; keep it private end to end on your own Matrix server.

The media skill

One download teaches your agent to render images, music, voice and video on the same key.

Questions

Can I run OpenClaw or Hermes Agent on a private model?

Yes. Add legit2compute as an OpenAI-compatible provider (base URL https://api.legit2compute.com/nodes/llm/v1, your key, a model with tool calling), then switch on Private in your developer dashboard. From then on every request your agent makes is served only by machines we own and operate. Step-by-step: OpenClaw, Hermes Agent.

Do you store my prompts or use them for training?

No. We record metadata only: the model, token counts, latency and status. Prompts and outputs are not stored, so there is nothing to train on, with Private on or off.

What changes when Private is on?

Your requests are served only by machines we own and operate: never a community provider's node and never an outside service, with no fallback to either. Every response carries X-L2C-Privacy saying which tier served it. Private requests bill at 1.5× the listed token price.

What if none of your machines has the model loaded?

A private request waits for it: you get 503 with "warming": true and a Retry-After, never a quieter route somewhere else. Models marked always on are kept loaded, so an agent pointed at one does not meet that.

Can I try it without an account?

Yes. The agent guides carry a public demo key that calls ornith-1.5:9b, free, shared by everyone and rate-limited. It runs on the shared tier; Private needs your own key.

Which models suit agents?

Ones with native tool calling and at least 64K of context, listed on this page from the live catalog. Pick one marked always on: it is kept loaded, so your agent never waits for it to start.

More on the OpenAI-compatible API and the docs.

Give your agent a model you can trust with its data.

Sign up with Discord, Google, GitHub or Twitch, copy your key, switch on Private.