Bring any client
OpenAI Chat Completions, the Responses API, and Anthropic Messages all land on the same gateway. No SDK changes.
Live gateway · — models online
Point your OpenAI or Anthropic client at a single URL and route to any model behind it. Automatic failover, per-token billing, and one key to rotate — not forty.
curl https://…/v1/chat/completions \
-H "Authorization: Bearer br_…" \
-d '{"model":"deepseek-v4.1-flash",
"messages":[{"role":"user",
"content":"Hello"}]}'
No SDK to learn, no proxy to run, no provider keys in your app.
Sign up and mint a key. Attach the models it may reach, or leave it open to everything published.
Change one URL. Existing OpenAI and Anthropic SDKs work unchanged — just swap the base URL and key.
Failover, retries and per-token accounting happen at the gateway. Watch spend and latency in the portal.
Before
# a key per provider, a client per SDK OPENAI_API_KEY=sk-… ANTHROPIC_API_KEY=sk-ant-… # plus retry logic, rate limits, cost maths
After
base_url = "https://…/v1" api_key = "br_…" model = "your-combo" # routing is ours
Combos route across the providers below with failover, so one dead endpoint never becomes your outage.
The gateway handles the failure modes, the metering and the keys.
OpenAI Chat Completions, the Responses API, and Anthropic Messages all land on the same gateway. No SDK changes.
Bundle models into a combo with fallback, round-robin, weighted or fusion strategies. A failing endpoint is skipped, not surfaced.
A credit wallet priced per model, with spend, tokens and latency logged per request and per key.
Scope a key to specific models, set request and token caps, rotate it in one click, and see what each key spent.
Cache breakpoints survive the hop, and cached tokens are billed at their own rate — a cache read costs a fraction of fresh input — so the saving is visible on every request.
Filter the request log by model, key and outcome, then export the whole filtered history as CSV for your own books.
No subscription, no per-seat fee, no minimum. Prices are per million tokens and shown up front.
Every account starts with welcome credit. Run out and paid models pause — free models keep working, and nothing is charged without a balance.
Credit is prepaid and non-refundable, including unspent balance. This gateway is a personal project, not a company — see the Terms.
Usually one line. Set the base URL to your gateway and use a br_… key. OpenAI and Anthropic SDKs work as they are; if you call the REST API directly, the request shape is unchanged.
A combo lists several models. If one fails, the request is retried against the next before your client ever sees an error, and the failed endpoint is cooled down rather than retried into the ground.
Per token, at the price shown for each model — input and output priced separately. Spend is written to your ledger per request, and the portal shows it by model, by key and by day.
Yes. Scope a key to specific models and set request or token caps. Each key's spend is tracked separately, so a runaway script is identifiable from the portal.
Requests are logged as metadata — model, tokens, status, latency, cost. Prompt and completion text is not retained. See the privacy policy.
Request a top-up from the Wallet tab. The operator reviews it and your balance is credited; you can also set a low-balance email alert so it never runs dry unannounced.
No. lbrouter is run by one individual as a personal project — no staff, no support desk, no service-level commitment. It is offered as-is because it is useful, not as a business. Questions go to admin@lbrouter.com.
No. Credit is prepaid and non-refundable, including an unspent balance and any credit left when an account is closed. Full detail is in the Terms.
Create an account, mint a key, change one URL.
Sign in to manage your keys and credit.
New here?