Build something
without the limits.

One API key. Your choice of six models. Familiar APIs that fit right into your workflow.

Choose your endpoint

One address for the website and API.

The website, documentation and dashboard use https://manylane.chat. Connect API clients to the same host using the base for their protocol below. No separate API subdomain is required.

01https://manylane.chat/v1

01https://manylane.chat

The Messages SDK adds /v1/messages itself: its base must not end in /v1. Direct HTTP requests use the full endpoints listed above. The tool guides below already use the correct base for each client, including Cursor aliases.

One account code. Separate API keys.

Your account code signs you into the website. It is never accepted as an API key. In API keys, create a named key for each application. The full key is displayed once; copy it into your application's secret manager.

01MANYLANE_API_KEY=vl_sk_your_generated_secret_key

The website uses an HttpOnly session cookie. API routes use explicit API keys only, so a browser session cannot accidentally authorize third-party API requests.

Your plan, your pace.

Membership is $5 for 24 hours, $10 for 7 days, or $30 for 30 days. Every plan includes the same catalog and unlimited requests over the membership period, subject to the disclosed technical and acceptable-use limits. There are no automatic renewals.

Open Membership, select your plan, accept the purchase terms, then prepare a checkout. When payments are enabled, the server returns a unique order, the exact USDT quote and expiry, and a secure checkout link. Send USDT on BNB Smart Chain (BEP20) only. Never send funds while checkout is disabled or use an address from an unsolicited message.

Use a blockchain withdrawal, not Binance Pay or a Binance internal transfer. The full quoted amount must arrive after any withdrawal fees. Ethereum (ERC20), Tron (TRC20), and Base are not valid networks for a USDT/BEP20 quote. Account registration remains code-only: no email, personal name, or connected exchange account is required by Manylane.

Selecting a plan does not charge you. Returning from checkout, a wallet screenshot, or a transaction submitted for confirmation does not activate a plan. Manylane checks the payment service directly and activates each verified paid order once. Existing membership time is preserved: a new period starts at your current expiry if it is still in the future.

The page checks open orders while visible and online; server checks continue independently. For a partial payment, check the remaining balance in the original checkout before sending more. Do not resend the original amount. Expired, late, refunded, or reversed payments may require operator review. Keep your order ID for support, never share your account code, and read the purchase terms before paying.

Exact model IDs

Use these IDs exactly. Manylane never silently substitutes another model. Catalog inclusion is separate from live availability; check service status and test the capabilities your application needs.

Use a non-empty messages array with a model ID. Standard text messages, streaming, and function tools are forwarded to the selected model. Model support for images, structured output, reasoning settings, and tools can differ.

01import OpenAI from 'openai';0203const client = new OpenAI({04  apiKey: process.env.MANYLANE_API_KEY,05  baseURL: 'https://manylane.chat/v1',06});0708const response = await client.chat.completions.create({09  model: 'qwen3.8-max-0902',10  messages: [{ role: 'user', content: 'Think bigger.' }],11});12console.log(response.choices[0].message.content);

Send a string or input-item array in input. This release supports stateless text responses, streaming, and function tools. Keep conversation history in your application and resend it in input.

01import OpenAI from 'openai';0203const client = new OpenAI({04  apiKey: process.env.MANYLANE_API_KEY,05  baseURL: 'https://manylane.chat/v1',06});0708const response = await client.responses.create({09  model: 'qwen3.8-max-0902',10  input: 'Think bigger.',11  store: false,12});13console.log(response.output_text);

Use the Anthropic SDK with your Manylane origin as baseURL. Include max_tokens and messages. System instructions belong in the top-level system field. We preserve the native Messages response and event formats.

01import Anthropic from '@anthropic-ai/sdk';0203const client = new Anthropic({04  apiKey: process.env.MANYLANE_API_KEY,05  baseURL: 'https://manylane.chat',06});0708const message = await client.messages.create({09  model: 'qwen3.8-max-0902',10  max_tokens: 1024,11  messages: [{ role: 'user', content: 'Think bigger.' }],12});13console.log(message.content);

Compatible protocol does not mean access to Claude models. Choose one of the six catalog IDs above. Send anthropic-version: 2023-06-01 when using HTTP directly.

Stream, don't wait.

Set stream: true to receive Server-Sent Events. Streaming is incremental, not a buffered response disguised as streaming. Closing the client stream cancels the upstream request and releases the account's concurrency slot.

01const stream = await client.chat.completions.create({02  model: 'qwen3.8-flash',03  messages: [{ role: 'user', content: 'Hello.' }],04  stream: true,05});0607for await (const chunk of stream) {08  process.stdout.write(chunk.choices[0]?.delta?.content ?? '');09}

For Chat Completions we request final usage metadata automatically. Token counts appear in the dashboard when the model reports them; missing usage is marked as unreported, not estimated. An HTTP 200 can still be followed by an SSE error, so check event errors and the terminal event.

Input, output and cache tokens.

The dashboard shows model-reported usage, not an estimate or a per-token bill. Open Usage & activity to view the breakdown by period and model, or expand an individual request.

01{02  "input_tokens": 100,03  "output_tokens": 20,04  "cache_read_tokens": 60,05  "cache_write_tokens": 10,06  "input_tokens_reported": 1,07  "output_tokens_reported": 108}

In this example, total usage is 120, not 190. A cache-read rate of 60% applies to the known input. Coverage counts show how many attempts actually reported each field, including partial or interrupted requests.

API field mapping

Raw inference responses keep their native protocol schema. Normalization is for Manylane’s usage ledger. Streaming usage events replace cumulative counters instead of adding them repeatedly. A disconnected stream can have partial counts and is not recorded as a successful completion.

CSV exports append the four detail/reporting columns; missing fields are empty cells. JSON exports use null. Per-field reporting flags are 1 for reported, 0 for missing, and null for legacy records. No model is promised to expose cache metadata, a specific hit rate, or a particular cache lifetime.

Unlimited requests. Clear boundaries.

There is no per-day request count or membership token budget. Technical limits keep access usable: 2 simultaneous requests per account, a 1 MiB request body, a five-minute request deadline, one completion per request, and the selected model's context and output limits.

Every response includes an X-Request-ID for troubleshooting. Errors intentionally omit provider credentials and internal diagnostics. Failed or interrupted inference attempts remain visible in your request log.

Useful insights, without a transcript.

Manylane stores request metadata: model, protocol, status, reported token counts, timestamps, and duration. It does not store your prompts or generated text in its application database. Your content is sent to a third-party inference service to generate a response; its processing policies apply.

Playground conversations live only in the current browser tab. API keys and account codes are stored as cryptographic hashes. Request metadata is retained for up to 90 days, with daily cleanup. You can export metadata or delete your account in Settings.