prepaid billing for AI-orchestrated work

You orchestrate the build.
LLM Central bills it.

Every AI build session is priced — usage and compute — debited from prepaid balance and turned into a clean receipt. Claude Code, Codex, VS Code, GitHub Copilot or any MCP client: no spreadsheets, no guessing what an AI build costs.

Verde Café — Loyalty app
session cc-8842 · billed
auto-debited
Compute 1.63M tokens · ×1.30$18.46
MCP tools 42 calls$1.90
Infra allocated$3.20
Session charge$23.56
balance $781.00 → $757.44top-up $200.00
Client sees: “Loyalty app · $23.56 · balance $757.44.”
to the cent
Every balance reconciles to an append-only ledger
4 token classes
Input, output, cache-write & cache-read all priced
0 jargon
Clients see value and balance — never tokens

How it works

You’re the gateway — orchestrating models, MCP tools, and infrastructure. LLM Central turns that work into a bill, automatically.

01

Orchestrate

You direct Claude Code across models, MCP tools, and infrastructure to build the app. You are the gateway.

02

Meter

When the session ends, a hook posts its usage. LLM Central prices the compute (four token classes, incl. cache).

03

Bill

It debits the prepaid balance through the ledger, fires an auto-top-up if needed, and issues the client a plain-language receipt.

Built for the way AI work is priced now

Hourly billing breaks when AI collapses hours. LLM Central prices the orchestrated session instead — transparently to you, simply to the client.

Prepaid wallets + auto-top-up

Clients fund a balance up front. Each session debits it, and it refills automatically before it runs dry — billing never blocks a finished build.

Receipts, not token dumps

Clients see “Loyalty app · $23.56 · balance $757.44.” Every receipt is signed; the token-level detail is your private audit trail.

An append-only ledger

Every debit, top-up, and adjustment is one immutable entry. Balances derive from the ledger and reconcile to the cent — nothing is untraceable.

It bills itself

A one-line Claude Code hook posts each finished session. Pricing, settling, and receipts happen with no manual step.

Connect in one hook

Add a SessionEnd hook to Claude Code and set the project you’re billing. Finished sessions post themselves — priced, settled, and receipted. Idempotent, so nothing double-charges.

  • Parses tokens (all four classes), tool calls, and duration
  • Never blocks Claude Code — queues offline, retries
  • Hours auto-fill from duration; override any session in the console
~/.claude/settings.json
{
  "env": {
    "LLMCENTRAL_URL": "https://llm-central.com",
    "LLMCENTRAL_INGEST_TOKEN": "•••••••••",
    "LLMCENTRAL_PROJECT": "verde-cafe-loyalty"
  },
  "hooks": {
    "SessionEnd": [
      { "hooks": [
        { "type": "command",
          "command": "node emit-session.mjs",
          "timeout": 60 }
      ] }
    ]
  }
}
LLM Gateway

A gateway your clients call directly

Beyond billing the builds you orchestrate, LLM Central fronts the models and MCP tools themselves. Issue each client an API key — with identity, access control, and full request logging — and every call meters into the same prepaid wallet, ledger, and receipt.

One API, many models
Route to any provider's models — and your MCP tools — through a single endpoint.
Per-client API keys
Identity and access control built in; scope and revoke per client.
Full request logging
Every request, token, and cost recorded and attributable.
Same wallet & receipt
Gateway usage debits the prepaid balance and reconciles to the ledger.
client.py
from openai import OpenAI

client = OpenAI(
    api_key="llmc_your_client_key",
    base_url="https://api.llm-central.com/v1",
)

# any model + your MCP tools,
# metered to the client's wallet
resp = client.chat.completions.create(
    model="claude-opus-4-8",
    messages=[{"role": "user", "content": "Ship it."}],
)
GPU rental

Rent GPU + infrastructure by the hour

Shared GPU compute for open-source models — spin up on our infrastructure and pay only for the hours you use. Live now, from $0.05 to $4.00/hr across the fleet, metered into the same prepaid wallet and receipt.

RTX 3090
Ampere24GB VRAM
from$0.07/hr
median $0.15/hr
Rent
RTX 4090
Ada Lovelace24GB VRAM
from$0.13/hr
median $0.37/hr
Rent

More GPUs and custom configurations available. For rates and provisioning, contact sales at [email protected].

Money you can audit

Balances aren’t a number someone typed. They’re derived from an append-only ledger, written inside serializable transactions, idempotent on every gateway event. Each receipt is signed. Reconciliation is a subtraction, not a leap of faith.

Append-only ledger
Debits, top-ups, adjustments — one immutable row each.
Idempotent
Replayed sessions and webhooks never double-charge or double-credit.
Signed receipts
Tamper-evident, with a jargon-free view for the client.
Auto-top-up
Refills at a threshold — one pending refill per client, never a pile-up.

Start billing your builds

Onboard a client, point the hook at their project, and finish a session. It bills itself — you watch it land.

Open the console