Skip to content

The basics

What is Smouter?

Smouter is an OpenAI- and Anthropic-compatible AI gateway: one API key, 50+ models and one prepaid wallet. You point your existing SDK or coding tool at Smouter’s base URL, keep your code, and every request is bought from a benchmarked pool of providers serving the same model. Chat, images, audio, embeddings, web search, batches and MCP tools all run on the same key. New to the idea? Read what an LLM gateway is.

How is Smouter cheaper than calling the model provider directly?

Each request is shopped across a pool of providers serving the same model weights and bought at the lowest price that actually holds, re-quoted on every call. On top of that, response caching is on by default and you can opt into auto routing. Value-lane prices run up to 50% under OpenRouter on the same model, though savings vary by model and workload. See live prices on the models page and how to reduce LLM API costs.

Is it the same model, or a cheaper substitute?

It’s the same model. Every node in the pool is fingerprinted against the reference weights and benchmarked for latency and error rate continuously, not just at onboarding. Nodes whose weights drift, or that advertise a low price and fail to serve, are demoted. If you need the model maker’s own endpoint, send the request on the verified lane instead.

Who is Smouter for?

Developers and teams who call LLM APIs from their own code, and people who run coding agents like Claude Code, Cursor or Cline and want one key for every model. It also suits app builders who want their users to pay for their own inference through Connect with Smouter. If you already use an OpenAI-compatible client, switching is a one-line change. See OpenRouter alternatives for how gateways compare.

Which models can I use through Smouter?

More than 50 models across chat, images, speech, transcription and embeddings, including open-weight models like Kimi K3, DeepSeek V4, GLM-5, Qwen3 Coder and gpt-oss-120b. Availability is computed from live capacity, so the models page and the public GET /v1/models endpoint are the source of truth. See cheap open-weight model APIs for more.

Using the API

What is the Smouter API base URL?

For OpenAI-style clients the base URL is https://api.smouter.ai/v1. The /v1 suffix matters, and leaving it off is the most common setup failure. Anthropic-style clients such as Claude Code use https://api.smouter.ai with no suffix, because they append /v1/messages themselves. Authenticate with an sk-smouter-… key in the Authorization: Bearer header. The quickstart has copy-paste examples.

Can I use the OpenAI SDK with Smouter?

Yes. Smouter is OpenAI-compatible, so the official OpenAI SDKs and frameworks like the Vercel AI SDK or LangChain work unchanged: set the base URL to https://api.smouter.ai/v1, pass your Smouter key and use a model id from the catalog. Chat completions, the Responses API, embeddings, images and audio all follow the OpenAI shapes. See what an OpenAI-compatible API is and the any OpenAI client guide.

Does Smouter support the Anthropic Messages API?

Yes. Smouter serves POST /v1/messages natively, including streaming, tool use and system prompts, so Anthropic SDKs and Claude Code work by changing the base URL to https://api.smouter.ai. Keys are accepted as a bearer token or in the Anthropic-style x-api-key header. Auxiliary Anthropic endpoints such as token counting and files aren’t proxied. The Claude Code guide has the exact setup.

Can I use Claude Code, Cursor or other coding agents with Smouter?

Yes. There are step-by-step setup guides for Claude Code, Cursor, OpenCode, Codex CLI, Cline, Roo Code, Kilo Code, Continue, Aider, Zed and desktop chat apps. Each reduces to a base URL, a key and a model id. See using Claude Code with other models.

Where do I find the right model id?

Copy it from the models page or call GET /v1/models, which needs no authentication. Ids are matched verbatim and are case-sensitive, with no vendor prefixes and no renaming. A mismatched id returns 404 model_not_found on most tools. Each catalog row also carries a kind and pricing_unit, so you can tell chat, image, embedding and audio models apart programmatically.

What does model auto do?

Send model: "auto" and Smouter resolves the request to one concrete available model, the cheapest one that can handle it, before pricing. You don’t pin anything, and you can always name a model yourself to skip this step. Auto requests aren’t cached, and keys scoped with a model allowlist can’t use it. Read more in LLM model routing.

How do I set fallback models?

Send a models: [...] list in the request body, or the x-smouter-models header when the body field is absent, with up to 16 ids. Smouter picks the first viable candidate at dispatch. It isn’t a retry after a model has started generating. Every candidate must pass your key’s allowlists, and the list is never forwarded upstream. See LLM API fallbacks and the docs.

Does Smouter support streaming, tool calls and JSON mode?

Yes. Chat completions support streaming with stream: true (standard server-sent events), tool calls and structured output when the chosen model supports them, and the Responses and Messages endpoints stream too. For a live stream, the final charge is settled after the response drains and shows up in Usage. A few features, such as programmable routes and output guardrails in blocking mode, need non-streaming calls. Details in the API reference.

Is there a batch API for async jobs?

Yes. POST /v1/batches takes chat-completion lines, runs them asynchronously within a 24-hour window and returns results as NDJSON. Completed batches earn a discount, and the current rate is shown in the dashboard as a ceiling, since the realized credit can be lower on some models. Batch isn’t available on zero-retention keys. See batch processing and the LLM batch API guide.

Pricing and billing

How does Smouter pricing work?

You pay per token at live rates, from prepaid credits. There is no subscription, no minimum and no seat fee for API use. Images are billed per image, speech per character and transcription per second of audio. Because prices follow live supply, they aren’t fixed on a rate card: check live prices on the models page. For how token billing works in general, see LLM API pricing explained.

How do credits and top-ups work?

Smouter is prepaid: you add credits in Billing, and every request draws from that balance. Your full ledger of top-ups, charges and bonuses sits on the same page, so every dollar traces to a request or a payment. You can set a balance alert email so requests don’t start failing unexpectedly. See credits and top-ups for the details.

Can I try Smouter for free?

Yes. You don’t need a card to start, and your first credit is free, so you can make real calls before adding money. Nothing renews on its own. Create an account, verify your email, mint a key and follow the quickstart. When you need more, you top up credits and keep paying per token at live rates.

Can Smouter top up my balance automatically?

Yes. In Billing you choose the balance to refill at, how much to add and a monthly limit you never want to exceed, then save a card on Stripe’s secure page. Smouter never sees or stores your card number. Every automatic top-up is emailed to you, as is any failed attempt. Organization admins can set the same for the shared wallet. See credits and top-ups.

How do I see what each request cost?

Every response carries receipt headers when the values are known: x-smouter-charge for what you paid, x-smouter-baseline-openrouter and x-smouter-saved for the comparison, plus x-smouter-lane and x-smouter-cache. The Usage page shows the same per request, with filters by model, lane, key, tag and time range. See understand your usage.

Can I set spending limits on an API key?

Yes. Each key can have a monthly spend cap, a daily cap and a warning threshold, set independently. Keys can also carry model and provider allowlists, an expiry and a tokens-per-minute limit. A request that would exceed a cap is refused with 402 per_key_cap_exceeded before the model is called. See create and manage API keys.

Can my team share one Smouter account?

Yes, through organizations. You invite members by email, assign roles, and manage org API keys and usage in one place, and larger teams can configure single sign-on. An owner can switch the organization to a shared wallet so org keys bill the org instead of each member. See organizations, members and roles. For volume questions, contact us.

Does Smouter have a monthly subscription?

Subscription plans are rolling out and aren’t generally available yet. Today, the way to use Smouter is pay per token through the API: prepaid credits, no subscription, no minimum and no seats, open to everyone. The pricing page shows the current options, and live per-model prices are on the models page. For team or volume questions, talk to us.

Lanes, privacy and data

What is the difference between the value and verified lanes?

Value is the default: the sourcing pool of probed providers serving the same weights, where the discounts are and caching applies. Verified sends the request to the model maker’s own endpoint under a zero-retention contract, with nothing stored or cached. Choose per request with x-smouter-lane: value or verified. Not every model is on both lanes, so check the models page. See value vs verified lanes.

Does Smouter offer zero data retention?

Yes. Turn it on for a key, or per request with x-smouter-zero-retention: on. Prompts and responses are then not retained for replay or caching, while usage and billing metadata is still metered. The verified lane is zero-retention by contract and responds with cache-control: no-store. Batch jobs can’t run on zero-retention keys. See zero data retention LLM APIs and the data processing page.

What does Smouter store about my requests?

Usage records hold the model, outcome, token or unit counts, charge, attribution metadata and timestamps, not full prompt or completion bodies. Eligible responses may be kept for account-isolated caching and idempotent replay unless zero retention is on. Your metadata, account ids and Smouter credentials are never forwarded to the provider. The data processing page and privacy policy spell out the rest.

How does response caching work?

Caching is on by default and account-isolated. Exact response caching applies only to deterministic requests, meaning an explicit temperature: 0 or a pinned seed, and prefix savings come from upstream warm-prefix accounting. Semantic caching is opt-in with x-smouter-cache-semantic: on. The x-smouter-cache header on each response shows the result. See how response caching works and LLM caching explained.

How do I turn caching off for a request?

Send x-smouter-cache: no-store to skip both the cache lookup and storage, or x-smouter-cache: no-cache to skip the lookup but still store the fresh answer. You can also set a reuse window with x-smouter-cache-ttl and partition entries with x-smouter-cache-namespace. Zero retention and the verified lane disable caching entirely. See how caching works and the docs.

Reliability and limits

What happens if a provider goes down?

The gateway fails over automatically when a node degrades, and nodes are health-checked and benchmarked continuously, so a single bad provider shouldn’t reach you. If no healthy capacity is left for a model, you get 503 service_unavailable, sometimes with retry-after. You can also list fallback models so a different model is chosen at dispatch. See LLM API fallbacks and rate limits and TPM caps.

What are Smouter’s rate limits?

Requests can be limited at two levels: platform limits that protect the gateway, and limits you set yourself, including tokens-per-minute caps per key or as an aggregate ceiling for an organization. A limited request returns HTTP 429 with retry-after, so back off and retry. If steady traffic hits limits, open a ticket with your throughput profile. See rate limits and TPM caps.

Is Smouter down right now?

Check the status page. It probes the gateway API and routing live from your browser and lists current incidents and maintenance. If you see errors that the status page doesn’t explain, open a ticket or email support@smouter.ai with the request id, model, lane and a timestamp. See reporting an incident.

How do I contact Smouter support?

Open a ticket from your dashboard, which is the fastest route, or email support@smouter.ai. Both go to the same team and replies arrive by email. For a failed request or a charge that looks wrong, include the request id, model, lane and a timestamp from the Usage page. The support center has help articles, and security issues go through responsible disclosure. You can also use the contact page.

Didn’t find it? Search the support center or contact us.