llms — multi-account LLM routes, one URL

Product

A hosted multi-account LLM hub.

llms is operated by goodtek. Provider accounts, quotas, and failover live on the server. Callers see one OpenAI- or Anthropic-compatible endpoint. This is a product you attach to—not a self-hosted stack.

01

Connect accounts

Register OAuth or API keys with llms add. Secrets stay in the server vault.
02

Publish a route

Pick a preset and get a stable URL. Change policy later without touching client settings.
03

Attach any tool

IDE, CLI, bot, or script—set Base URL and sk-gt and you are done.

Routing

Routing presets you can actually pick.

Choose one of five presets in llms route create. Before a 429 or auth failure commits, we buffer to TTFT then move to the next account.

failover

sequential

Try accounts in order. If #1 is blocked, use #2. The common work/personal backup pattern.

Recommended

balance

round_robin

Rotate accounts per request. Even burn-down when limits are similar.

Even spread

prefer-primary

weighted 80/20

Use the primary more often, secondary occasionally. Soft preference, not hard backup.

Weighted

quota-first

quota_aware

Prefer accounts with higher remaining %. Fits multi-subscription pools like Codex.

Live in use

parallel

race · API-key only

Race healthy API-key accounts; keep the first TTFT. OAuth/subscription accounts excluded by default.

Opt-in

Live example routes: codex-quota-first, openrouter-direct · URL shape /r/<slug>/v1

Quota

Remaining quota on the board.

After connect, home and status show per-account remaining % and reset time. Check headroom before routing; exhausted accounts show as cooldown.

How to check

Cached values: llms status. Re-read provider usage: --refresh.

Codex / Claude OAuth — provider usage APIs fill quota_remaining_pct and quota_reset_at
API-key accounts — soft-cap usage heuristic %
Cooldown — 429s appear on the board and are skipped briefly from the pool

$ llms status --refresh
codex:goodtek ● ok quota 91% ████████░░ reset ~5d
codex:technfin ● ok quota 62% ██████░░░░ reset ~5d

status board

WHAT YOU SEE

health ● / ○ / ✖ ok · cooldown · error

quota bar 0–100% remaining headroom

reset ~Nd provider window

routes strategy slug → /r/…/v1

Attach

Everything a client needs to know.

You do not need to know which app you will use next. Any tool that accepts an OpenAI-compatible (or Anthropic Messages) Base URL works the same.

Base URL https://llms.goodtek.xyz/r/<slug>/v1 API Key sk-gt-… (llms route key)

Human install: install · Machines: /install.md · /install.sh

Install guide install.md

Pricing

Trial once. Then Starter or Pro.

You pay for the hub—routing, quota board, route URL, and sk-gt. Upstream LLM spend stays on your own accounts. Card via Polar; USDT/JPYC via Unifi Pay.

Trial

$0 · 7 days, once

3 accounts · 2 routes · 2 keys
40 RPM · 5M soft tokens

Start trial

Starter

$5 / month

5 accounts · 3 routes · 3 keys
60 RPM · 10M soft tokens

Subscribe

Pro

$9 / month

20 accounts · 15 routes · 10 keys
120 RPM · 50M soft tokens

Subscribe

FAQ

Frequently asked questions

Written so search and answer engines can cite cleanly.

What is llms?

A hosted multi-account LLM hub by goodtek. Provider accounts stay on the server; clients only need a route URL and sk-gt.

Which routing presets exist?

failover (sequential), balance (round_robin), prefer-primary (weighted 80/20), quota-first (quota_aware), and parallel (API-key race).

How do I check remaining quota?

Use llms status for cached values, or llms status --refresh to re-read Codex/Claude OAuth usage.

Which clients work?

Any tool that accepts an OpenAI-compatible or Anthropic Messages Base URL — IDEs, CLIs, bots, scripts. Not tied to one app.

Do I need to self-host?

No. The product is hosted at https://llms.goodtek.xyz.

Many accounts,one address.

A hosted multi-account LLM hub.

Connect accounts

Publish a route

Attach any tool