Direct answers to the questions developers actually ask before signing up. Grouped by topic. If your question isn't here, email support@stormfire.io or ask in Discord.
Change two things in your OpenAI SDK initialization: the api_key to your Stormfire key, and the base_url to https://api.stormfire.io/v1. Nothing else changes. Streaming, function calls, tools, vision, structured outputs, embeddings, audio, image generation all pass through transparently. If your code uses environment variables (OPENAI_API_KEY, OPENAI_BASE_URL), just update those values and don't touch the code at all.
Yes, fully. The tools parameter, tool_choice, parallel tool calls, and structured-output JSON mode all pass through to the upstream provider. Same for Anthropic's tool_use blocks when you call Claude models. There's no translation layer or compatibility shim — what you send is what reaches the upstream, and what they return is what you get.
Yes. Set stream=True in the OpenAI SDK and you'll get server-sent events back identically to calling OpenAI direct. We don't buffer responses. First-token latency from Tokyo egress is typically 250–500ms for GPT-4o, 200–400ms for Claude Sonnet, 150–300ms for Gemini Flash, depending on the upstream's current load.
The OpenAI SDK is the recommended path because it's the unified interface across all 30+ models. We also expose Anthropic-native endpoints at /v1/messages for users who want native Claude SDK compatibility. Google's SDK is OpenAI-compatible already in their Gemini 1.5+ releases, so it works through our standard endpoint.
All work out of the box. In LangChain, instantiate ChatOpenAI with our base_url. In LiteLLM, set api_base in your config. In LlamaIndex, use the OpenAI LLM class. CrewAI uses LiteLLM under the hood so it Just Works. We've never seen a framework break against our endpoint because we don't add any non-standard behavior.
Call GET /v1/models with your API key. Returns the OpenAI-format model list including every model you have access to. The dashboard also shows the same list with pricing inline.
Canonical upstream names. gpt-4o, gpt-4o-mini, claude-3-7-sonnet-20250219, gemini-2.0-pro, deepseek-chat, mistral-large-latest. No provider prefixes (unlike OpenRouter's anthropic/claude-3.7-sonnet). The /v1/models endpoint shows the exact accepted strings.
No, and that's intentional. The OpenAI SDK is already shipped, tested, and maintained in every language you care about. Building a Stormfire-branded wrapper would just be lock-in for no benefit. Use the official OpenAI library and you'll always be on the latest features.
$15 USDT. We picked $15 because below that, the network fee on a withdrawal (especially from Binance) becomes a meaningful percentage of the top-up. $15 with a $1 TRC20 withdrawal fee is a clean ratio.
TRC20 (Tron) by default — cheapest withdrawal fees on every major exchange ($1 typical), fastest confirmations, broadest support. BEP20 (BSC) if you're inside Binance and want $0.30 withdrawal. Polygon if you're already on Polygon for Web3 reasons. ERC20 only if you're already on Ethereum L1 and don't want to bridge — gas fees of $15–30 make small top-ups uneconomical.
TRC20: 1–2 minutes after broadcast. BEP20: ~30 seconds. Polygon: ~1 minute. ERC20: 2–5 minutes depending on gas. We credit your balance after one confirmation, which matches the standard practice for stablecoin operational use.
Three common causes. (1) Wrong chain — you selected TRC20 on the dashboard but withdrew on ERC20. We can recover this if the address happens to overlap, but it's slow; email support with the txhash. (2) Below minimum — sub-$15 deposits are queued, not credited; either top up the difference or email for refund. (3) Network congestion — rare on TRC20/BSC, normal on busy Ethereum days. Paste the txhash in the dashboard support widget and we'll trace it.
USDT only at this time. We may add USDC in Q3 2026. We deliberately don't support BTC or ETH because their volatility makes operational billing painful for both sides.
No. Stormfire is intentionally USDT-only. If you can use a card, OpenAI direct or OpenRouter's card path are smoother for you. Stormfire's value proposition is for developers whose cards don't work or who prefer crypto rails.
Yes, see the pricing page. Current promotion gives +2% on $50–99, +5% on $100–499, +8% on $500+. Stacks with VIP discounts. Bonuses are added to your balance as standard credit, no expiry.
Unused balance is refundable. Email support@stormfire.io from your account email with a USDT-TRC20 destination address. We process within 1 business day, minus the network fee (typically $1). Used credits aren't refundable — the upstream provider has already charged us for those tokens.
No. Signup is email-only, no ID verification. We hold an internal AML threshold for very large accounts ($50k+ cumulative top-ups) where we may ask basic questions, but that's well outside the scope of a typical developer use case.
We keep request metadata (model name, token counts, your API key ID, timestamp, success/failure) for 90 days for billing reconciliation and abuse detection. We do not persist request bodies (your prompts) or response bodies after the request completes. This is the default behavior for all accounts.
Yes, for enterprise accounts. Email sales@stormfire.io. The mode drops all metadata after billing aggregation and signs a per-customer DPA. Pricing is the same; there's just a setup conversation.
Our gateway runs in Tokyo (AWS ap-northeast-1). Requests are proxied to the upstream provider's region (OpenAI US, Anthropic US, Google US/EU). Your prompts cross our Tokyo edge, the upstream's region, and back. Latency from APAC users is excellent; from EU users, comparable to going direct.
We follow GDPR and CCPA principles for the data we hold (email, payment metadata, request metadata). Per-customer DPAs are available on enterprise. For prompts and responses that flow through to OpenAI / Anthropic / Google, the upstream's terms govern — we're a pass-through processor.
We're not HIPAA-certified and we don't sign BAAs. If you're processing PHI, work with OpenAI or Azure OpenAI direct under their HIPAA-eligible programs. We can't add a compliance posture we don't have.
Globally accessible. We don't proactively geo-block, but we comply with valid legal orders. If your local jurisdiction restricts your use of foreign SaaS, that's your responsibility to evaluate. Standard mainland-China mainland traffic patterns hit our gateway with elevated latency due to standard infrastructure; we don't operate special routing into mainland.
40% of net revenue from anyone who signs up via your referral link, for the lifetime of their account. Paid monthly in USDT-TRC20. No cap, no tier drop-off, no claw-back. You can sign up in the dashboard under "Referrals". Pays out around the 5th of each month for the previous month's activity. Minimum payout threshold: $10.
Yes. Email sales@stormfire.io. Enterprise gets: custom DPA, zero-retention mode, dedicated rate-limit pool, invoiced billing if you really need invoices instead of USDT receipts, SLA, named support contact. Pricing is the same per-token; the "enterprise" part is the contract, not a markup.
Each account supports multiple API keys, which you can scope to projects or environments. Team/org features (per-user audit logs, role-based access, SSO) are on the enterprise tier.
USDT top-up receipts list the USD-equivalent amount and serve as invoices for most jurisdictions. If your accountant insists on a formal fiat invoice with a tax ID, that's an enterprise feature.
VIP2 (15% off) is the public tier. Above that — say, $5k+/month sustained spend — talk to sales for custom pricing. We can pass through more of our volume rebate from upstreams at high enough scale.
Yes. The enterprise tier includes a white-label option where your customers see your brand and you handle the support relationship. Useful for AI SaaS that doesn't want to expose the underlying gateway.
We target 99.9% for the gateway itself. Upstream provider outages (OpenAI down, Anthropic down) are passed through — we can't fabricate availability we don't have. The dashboard status page shows per-model health in real time. When an upstream is degraded, we automatically surface that to your error responses with a clear message instead of silent timeouts.
Inherits from the upstream provider, scaled by your tier. Free tier: roughly OpenAI Tier 1 / Anthropic Tier 1 limits per upstream. VIP1: Tier 2-equivalent. VIP2: Tier 3-equivalent. Concrete numbers vary per model and change over time as we adjust our shared pool — the dashboard shows your live limit per model.
You get an OpenAI-format error response with our error code. Common cases: upstream rate limit (we surface as 429 with retry-after), upstream outage (503), invalid model (404), insufficient balance (402), invalid API key (401). Standard exponential backoff in your SDK handles all of these correctly.
Dashboard shows real-time per-key, per-model token consumption with daily/weekly/monthly graphs. CSV export available. There's also a /v1/usage endpoint that returns the same data programmatically for plugging into your own observability stack.
Discord for community questions and quick triage. Email support@stormfire.io for account-specific issues — target response time 4 hours during APAC business hours, 12 hours overnight. Enterprise gets a named contact and an SLA.
Yes — status.stormfire.io. Shows gateway health and per-upstream model status. Subscribe to incident notifications via RSS or email.
Founded by Zibin in mid-2026. Small bootstrapped team, no VC, no growth-at-all-costs runway pressure. Margins come from VIP-tier markup and upstream volume rebates, both stable. We're not optimizing for an exit; we're optimizing to run profitably and answer support tickets ourselves. The founder is reachable on Twitter/X and Discord — feel free to ask.