FAQ

TokSpan is a unified AI API gateway that gives you a single OpenAI-compatible endpoint to access 200+ models from 50+ providers — including OpenAI, Anthropic, Google, Meta, DeepSeek, and more.

Here's how it works: you send requests to TokSpan's API in standard OpenAI format. TokSpan routes your request to the appropriate upstream provider, translates the format if needed, handles the response, and returns it to you. From your perspective, you call one API — TokSpan manages model routing, load balancing, rate limiting, automatic failover, and cost tracking behind the scenes.

You only need a TokSpan account. We handle all upstream provider relationships on your behalf — no separate accounts with OpenAI, Anthropic, or Google required.

• One key, every model — Access GPT, Claude Opus, Gemini Pro, and DeepSeek with a single API key. No managing multiple billing accounts across different providers.
• No vendor lock-in — Switch models by changing one parameter. Your code stays the same regardless of which provider serves the request.
• Automatic failover — If a provider goes down, traffic auto-routes to a backup provider serving the same model. Zero manual intervention.
• Global reach — A global edge network serving developers worldwide.
• Cost visibility — Unified billing and real-time usage tracking across every model and provider in one dashboard.

Getting started takes about 5 minutes:

1. Sign up at api.tokspan.com/register
2. Add credits to your account via credit card, Apple Pay / Google Pay, or your local wallet
3. Create an API key from the Dashboard → API Keys page
4. Make your first call — just change base_url to https://api.tokspan.com/v1 and use your new key

If you already use the OpenAI SDK, you're 30 seconds away. See the Quickstart guide for code examples in Python, TypeScript, and curl.

We don't offer a permanent free tier — here's why that's actually good for you:

• No subsidized quality degradation — Free tiers at scale require cutting corners on infrastructure, support, or model access. We invest everything into reliability and model coverage.
• Pay only for what you use — No minimum commitment. Top up as little as you need. Most developers spend under $5 in their first month of testing.
• Every model, no restrictions — Your paid balance gives you immediate access to every frontier model on the platform — no "free tier only" model limitations.

New accounts may receive promotional starter credits from time to time. Check the Dashboard after signing up.

We provide access to 200+ models across all major categories: LLMs (GPT, Claude Opus, Gemini Pro, DeepSeek, Llama, Mistral), Vision & Multimodal, Embeddings & Reranking, and Speech & Audio (TTS + STT).

Browse the full live catalog with current per-token pricing at api.tokspan.com/pricing. See Models documentation for category overviews and capability comparisons.

TokSpan pricing is fully transparent. You pay for the tokens you consume, and every model's per-token price is shown up front.

Every model's per-token price is visible in real-time on the Model Hub before you make a call. The price you see is the price you pay — no hidden costs, no surprise bills.

There is no monthly subscription and no minimum commitment. Top up any amount and start calling models immediately.

We accept credit and debit cards (Visa, Mastercard, American Express) securely via Stripe, plus Apple Pay and Google Pay. We also support local wallet payments (e.g. Alipay and WeChat Pay) where available in your region — the exact methods shown at checkout depend on your location.

All payments are denominated in USD. Your balance updates immediately after payment is confirmed.

Yes, with the following policy:

• Card & wallet payments — Unused credits are eligible for a refund within 7 days of purchase. The platform service fee (the margin we add on top of provider rates) is non-refundable for tokens already consumed.

To request a refund, contact support@tokspan.com with your account email and payment details.

Yes. Unused credits expire 12 months after purchase. This policy allows us to manage provider-side liability while keeping our service fee as low as possible. You'll receive email reminders 30 days and 7 days before any credits expire. Expiration dates are visible in the Dashboard under Billing → Credits.

TokSpan notifies you before your balance runs out, so you can top up in time:

1. Set your low-balance alert threshold in Dashboard → Billing
2. When your balance drops below this threshold, you'll receive an email alert, and we recommend enabling the webhook / in-app notification for real-time alerts on important events
3. When you hit the threshold, simply top up via your card, Apple Pay / Google Pay, or a local wallet on the Billing page

For uninterrupted production workloads, keep your balance healthy and set alerts so you're notified before funds run out. You do not have to worry about surprise charges — requests stop cleanly with a clear insufficient_user_quota error when balance reaches zero, and you can top up whenever you are ready.

You have several layers of cost control:

• Per-key spending cap — Set a hard spending cap per API key. When the key exhausts its quota, requests are rejected until you raise the limit or top up.
• Platform rate limits — RPM (requests per minute) and TPM (tokens per minute) limits apply to your account and plan group; higher limits are available on request.
• Model whitelisting — Restrict a key to only specific models (e.g., only allow cheaper models for development keys).
• IP whitelisting — Limit key usage to specific IP addresses or ranges.
• Real-time usage dashboard — Monitor spend by key, model, and time period in the Dashboard.

Per-key limits (spending caps, model access, IP whitelist) are configured from Dashboard → API Keys when creating or editing a key. See Key Scoping for details.

Tokens are counted exactly as the upstream provider counts them: prompt tokens + completion tokens = total billable tokens. We pass through each provider's tokenizer so counts match what you'd see calling the provider directly.

For multimodal inputs (images, audio, video): each provider has its own formula for converting media to token equivalents. We pass through the provider's calculation — your dashboard shows the final token count and cost. Generally: images in high resolution are ~85–170 tokens each (GPT), audio is ~50 tokens per second (speech-to-text).

For models with prompt caching (Claude, GPT), cached prompt tokens are billed at the provider's discounted cache rate. Your usage dashboard breaks down regular vs. cached token spend.

No — we do not store your prompts or completions. Your data passes through our infrastructure in real-time and is not persisted. We log request metadata only (model, token count, timestamp, API key ID) for billing accuracy and abuse prevention. This is necessary to calculate your usage and protect the platform.

We never use your prompts or completions for model training, and we never sell or share your data with third parties.

This depends on the provider and your settings:

• OpenAI — Does not train on API calls (as of March 2023). Data sent through the API is not used for model training by default.
• Anthropic — Does not train on API data. Claude API data is not used for model improvement unless you explicitly opt in.
• Google (Gemini) — Does not train on paid API usage. Free-tier usage may be subject to different terms.
• Other providers — Policies vary. Check each provider's terms. We recommend reviewing the provider's data usage policy if this is a concern.

If this is a priority, our team can help you select providers with clear no-training policies for your workloads — contact support@tokspan.com.

All API keys are encrypted at rest (AES-256). You can view your keys again at any time in the Dashboard — if a key is lost or suspected compromised, revoke it and generate a new one. All API traffic is encrypted with TLS 1.3 in transit. We recommend using API keys with IP whitelisting, model scoping, and monthly spending caps for defense in depth. Rotate keys every 90 days as a security best practice.

Security is foundational to TokSpan. Our infrastructure is built around these controls:

• TLS 1.3 — all data in transit is encrypted end-to-end.
• AES-256 — API keys and credentials are encrypted at rest.
• MFA — multi-factor authentication supported on all accounts.
• GDPR-aligned practices — data minimization and privacy disclosures are described in our Privacy Policy.

Enterprise customers can request additional details about our security posture at support@tokspan.com.

Yes. Our hosted service implements GDPR and CCPA requirements: data minimization (we don't store your prompts/completions), purpose limitation (metadata is only used for billing and abuse prevention), user data deletion on request, and transparent data processing disclosures.

For strict data residency or compliance requirements, contact our team — we can provide a DPA (Data Processing Agreement) and discuss your needs. Email support@tokspan.com.

You can delete your account and all associated data from Dashboard → Settings → Account → Delete Account. Account deletion is permanent and irreversible.

Regarding your balance:

• Any remaining credits at the time of deletion are forfeited. We recommend using your balance or requesting a refund before initiating deletion.
• Billing records required by law (payment receipts, invoices) are retained for the statutory period — typically 7 years.
• All other account data (API keys, usage logs, email, settings) is permanently purged within 30 days.

Yes. DPAs are available for enterprise and team accounts. We provide a standard DPA that covers GDPR data processor obligations, subprocessor disclosures, cross-border transfer safeguards (SCCs), and breach notification commitments. Contact support@tokspan.com with your legal team's requirements — typical turnaround is 2–3 business days.

Yes, 100%. Change base_url to https://api.tokspan.com/v1 and your existing OpenAI SDK code works instantly — Python, Node.js, Go, curl, or any OpenAI-compatible library.

Supported features:

• Chat Completions — Full support including system/user/assistant messages
• Streaming — SSE (Server-Sent Events) with stream: true
• Function / Tool Calling — Parallel tool calls, strict mode
• JSON Mode — Structured output with response_format
• Vision / Image Input — URL and base64 images
• Embeddings — Full support across multiple providers
• Audio — TTS (text-to-speech) and STT (speech-to-text)
• Images — Generation via leading image models
• Rerank — Full support
• Reasoning Effort — Control reasoning depth via the reasoning_effort parameter on supported models

Some reasoning models accept the standard reasoning_effort parameter to control reasoning depth. TokSpan forwards this parameter to the upstream provider — support depends on the model itself:

Just include it in your request body — e.g. "reasoning_effort": "high". If a model doesn't accept the parameter, omit it and the model's default is used. See the model catalog for per-model support.

Switching models is a one-line change — just set the model field in your request to any model ID from the catalog:

• OpenAI-compatible: model: "MODEL_NAME" in the request body
• Anthropic-native: model: "MODEL_NAME" in the request body
• Gemini-compatible: model="MODEL_NAME" in generateContent

Your code stays the same — only the model name changes, and all formats share the same API key and balance. Browse the full catalog with live pricing at api.tokspan.com/pricing.

Yes. Prompt caching is automatically applied when the upstream provider supports it (Claude, GPT, Gemini). When you send the same prompt prefix across multiple requests, the provider caches the computation and charges you a discounted rate for the cached portion — typically 90% off the full prompt token price.

You don't need to change your code. TokSpan passes through the cache status from the provider, and your dashboard shows cached vs. uncached token usage. To maximize cache hits:

• Keep static content (system prompts, context documents) at the beginning of your prompt
• Put dynamic content (user messages, latest queries) at the end
• Reuse the same system prompt across requests

Read the full optimization guide: Prompt Caching.

TokSpan returns standard HTTP status codes:

We recommend implementing exponential backoff with jitter for 429 and 5xx errors. See Error Codes for the complete reference with error response bodies.

We target 99.99% uptime for our hosted API. Our infrastructure is horizontally scaled across multiple availability zones with no single point of failure.

Enterprise plans include a formal SLA with uptime commitments, credit penalties, and priority incident response. Contact support@tokspan.com for SLA terms.

TokSpan automatically fails over to the next healthy model. Our system:

1. Detects provider unavailability (5xx errors, timeouts) within seconds
2. Skips providers with recent outages (temporary cooldown window)
3. Routes to the next provider in the failover chain (managed automatically by TokSpan)

The chain is built across providers serving the same model (e.g., multiple providers serving the GPT series), ensuring zero dropped requests. You pay only for the provider that ultimately serves the request — failed attempts are not billed. See Auto Failover to learn how it works.

TokSpan adds minimal routing overhead — typically under 100ms for API requests. The dominant latency factor is the upstream model's generation speed (time-to-first-token and tokens-per-second), which is unaffected by routing through TokSpan.

For latency-sensitive applications:

• Enable streaming (stream: true) so users see tokens as they're generated
• Choose a faster or lighter model for interactive use — see the model catalog for latency profiles
• Route through TokSpan's global edge network — requests are served from the region nearest to your users

Yes. Check the API Status Page for real-time service health, incident history, and scheduled maintenance windows. Subscribe to incident notifications via email, webhook, Bark, or Gotify — see Webhooks for setup.

Log in to the Dashboard, go to API Keys, and click + New Key. Give each key a descriptive name (e.g., "production-backend", "mobile-app-dev", "staging-test"). You can create unlimited keys.

We strongly recommend creating separate keys for each environment (development, staging, production). Each key has independent usage tracking, spending limits, and model scoping — making it easy to monitor costs per environment or per client. If a development key is compromised, your production key is unaffected.

Revoke it immediately from the Dashboard — revocation takes effect instantly (under 60 seconds propagation). Generate a new key and update your application. Monitor the usage logs for any suspicious activity on the old key before deleting it.

Prevention best practices:

• Use IP whitelisting to restrict where keys can be used from
• Set monthly spending caps to limit blast radius if a key leaks
• Rotate keys every 90 days
• Never commit keys to version control — use environment variables or a secrets manager

Create a separate API key for each environment from the Dashboard:

• Development — Tight spending cap (e.g. $10), restrict to cheaper models, no IP restriction (for local dev)
• Staging — Moderate cap (e.g. $50), same models as production, IP-restricted to your staging servers
• Production — Higher cap as needed, all models, IP-restricted to your production servers

Each key has its own usage dashboard, alerts, and activity log. This setup gives you per-environment cost visibility and prevents a dev-phase bug from burning through your production budget. See Key Scoping for the full configuration reference.

Yes. Enterprise plans include:

• Dedicated account manager — Single point of contact for all questions
• Priority support — Guaranteed response times (typically under 2 hours for critical issues)
• Custom SLA — Uptime commitments with credit penalties tailored to your needs
• Consolidated invoicing — Net-30/60 terms, purchase orders, multi-currency billing
• Custom rate limits — RPM/TPM limits scaled to your workload
• Deployment assistance — Architecture review and integration guidance for your workloads

Contact support@tokspan.com with your requirements — we'll put together a proposal within 1 business day.

We offer multiple support channels:

• Email — support@tokspan.com for billing, account, and technical questions. We respond within 24 hours; most questions are resolved in a single exchange.
• Status Page — API Status for real-time incident updates. Subscribe for notifications during outages.

Enterprise customers receive a shared Slack channel with our engineering team for real-time collaboration.