FAQ
The questions people actually ask.
39 direct answers across pricing, models, SDK compatibility, privacy, and ops. Longer write-ups live in /blog and the setup guide.
Getting started
What is Anvat?
+
Anvat is an OpenAI- and Anthropic-compatible API gateway. Point one base URL (api.anvat.app) and one key at it, call any frontier model — Claude Opus 4.8, Sonnet 4.6, Haiku 4.5, GPT-5, Gemini 3.5 Pro, DeepSeek V4 Pro — and pay materially less than going direct.
How do I sign up?
+
Click 'Get $2 free' on anvat.app — email + password, no credit card. The $2 signup credit is enough to run several hundred Sonnet 4.6 requests end-to-end.
Do I need a credit card to try it?
+
No. New accounts get $2 in API credit on signup. You only attach Stripe when you want to top up or subscribe.
What's the fastest path from zero to first call?
+
Sign up, copy your API key from /app/keys, then paste a snippet from /snippets — most people are streaming tokens in under a minute. cURL works too, see /api-reference.
Can I use Anvat with Claude Code?
+
Yes. Set ANTHROPIC_BASE_URL=https://api.anvat.app and ANTHROPIC_API_KEY=<your-anvat-key>. claude-code routes through Anvat with no behaviour change — same models, same tool use, same MCP support.
Billing + pricing
How does pricing work?
+
Two ways to pay: metered tokens at 30% below published list rates, OR prepaid credit packs that double your spend (pay $100, receive $200 in API credit). Most teams combine both and pay roughly half the effective cost.
Are there subscription plans?
+
Yes. Pro, Pro x5, and Ultra monthly/yearly plans bundle a base credit allowance with priority routing and higher rate limits. See /pricing for the table.
What currencies are accepted?
+
USD via Stripe for all card and ACH payments. Crypto: USDC and USDT (TRC-20, ERC-20) for prepaid credit packs.
Do you charge by token or by call?
+
Per token, exactly. Anvat exposes the dollar cost of every call in the x-anvat-cost-usd response header and in your activity log — no rounding, no per-call surcharge.
Is the 30% discount available on every model?
+
Yes, on every Anthropic and OpenAI model in the catalog. Gemini, DeepSeek, and xAI tiers also ship with discounted metered rates, listed on /pricing.
Are there per-request minimums or hidden fees?
+
No. Anvat does not charge a per-call surcharge, a monthly minimum, or any setup fee. The bill is the metered token cost, full stop.
How does the 2× credit match work?
+
When you top up your wallet with a prepaid pack, the credit added to your balance is 2× the amount you paid. Pay $50 → receive $100. The credit doesn't expire and applies to every model in the catalog.
Can I see my spend in real time?
+
Yes. The dashboard at /app/usage shows token + dollar consumption per key, per model, per day. Each request emits an x-anvat-cost-usd response header so you can log it client-side too.
Models + capability
Which Claude models are available?
+
Claude Opus 4.8, Sonnet 4.6 (with 1M-context tier), and Haiku 4.5 — all the current public Anthropic frontier tiers. Per-model spec pages live under /models.
Are GPT-5, Gemini, and DeepSeek available too?
+
Yes. GPT-5 and GPT-5 Mini (OpenAI), Gemini 3.5 Pro and Gemini 3.5 Flash (Google), DeepSeek V4 Pro and DeepSeek V4 Flash (DeepSeek). One key, one base URL — see /models for the full catalog.
Do you support Claude prompt caching?
+
Fully. cache_control blocks pass through end-to-end. Cache writes (5m / 1h) and cache hits (0.1×) are billed at the same multipliers as direct Anthropic.
Do you support tool use / function calling?
+
Yes. Anvat preserves the tool-use contract byte-for-byte. Anthropic tool blocks and OpenAI tool/function calls both work; responses come back in the original shape.
Does Claude's MCP support work through Anvat?
+
Yes. MCP discovery, capability negotiation, and tool routing all round-trip through Anvat unchanged. See /mcp for the catalog of MCP servers and Claude Skills verified end-to-end.
Which model should I pick by default?
+
Sonnet 4.6 is the right default — frontier intelligence, half the price of Opus, plenty fast. Drop to Haiku 4.5 for cheap high-volume turns; jump to Opus 4.8 when the task is genuinely hard. See /benchmarks and /leaderboard for data.
Compatibility + SDKs
Which SDKs work with Anvat?
+
Every SDK that speaks the OpenAI or Anthropic protocol. Official Python + TypeScript SDKs for both vendors, plus LangChain, LlamaIndex, Vercel AI SDK, DSPy, LiteLLM. Full list with snippets at /integrations and /snippets.
Does Cursor work with Anvat?
+
Yes. In Cursor settings, set the OpenAI base URL to https://api.anvat.app/v1 and add your Anvat key. Then any model in the catalog works as a custom model. Setup guide at /use-cases/cursor.
Does Zed / Continue / Cline / Aider work?
+
Yes — see /integrations for copy-paste setup for each. Anvat is a base-URL change for any IDE that supports a custom OpenAI or Anthropic endpoint.
Can I migrate from OpenRouter without code changes?
+
Yes. Change https://openrouter.ai/api/v1 to https://api.anvat.app/v1 and swap the key — every OpenRouter-shaped client works. Comparison at /vs/openrouter.
Can I use Anvat as the upstream for my self-hosted LiteLLM?
+
Yes. Point your LiteLLM model_list config at api_base: https://api.anvat.app/v1 with your Anvat key. See the snippet at /snippets (filter to LiteLLM).
Performance + reliability
What's the gateway latency overhead?
+
Under 25 ms median (p50), under 80 ms p99, over a direct upstream call from the same region. Streaming TTFT is within 30 ms of upstream. SLOs published at /status.
Is the gateway streaming-capable?
+
Yes. SSE streaming is the default for /v1/messages and /v1/chat/completions. The stream is forwarded byte-for-byte — no buffering, no chunk repacking.
Are there rate limits?
+
Yes — per-plan RPM (requests per minute), concurrent request limits, and rolling 5-hour / 7-day spend caps to prevent runaway costs. Free tier is intentionally permissive enough to evaluate the product. Limits documented at /app/usage.
What's your uptime target?
+
99.9% monthly availability on the gateway. SLO table + browser-side health probe at /status (it's a real fetch from your network, not a pre-rendered badge).
Can I run a health check from my CI?
+
Yes. GET https://api.anvat.app/healthz returns 200 with a JSON body including uptimeSeconds and status when healthy. Plug it into Pingdom / your own check.
Privacy + security
Do you log prompts or completions?
+
No. Anvat only stores billing metadata: model, token counts, latency, cost, and request id. Prompt and completion bodies are forwarded to the upstream provider and never persisted on Anvat infrastructure. Operator-managed self-host can change this; the hosted product does not.
What data does Anvat see?
+
Anvat sees the request body in flight (required to route it). It is not written to disk and not retained. The upstream provider (Anthropic / OpenAI / etc.) handles the prompt per their own privacy policy.
Are you GDPR compliant?
+
Yes. EU data subjects can request export and deletion of their account data (email, billing history). Since prompts are not stored, there's no prompt content to export or delete.
Can I use Anvat for HIPAA / regulated workloads?
+
Not on the public hosted plan today. Regulated workloads should run a dedicated / self-hosted Anvat instance — get in touch at hello@anvat.app and we'll scope an option.
Where is Anvat infrastructure hosted?
+
Primary infrastructure runs on tier-1 cloud regions with redundant routing. Region pinning for upstream calls is on the public roadmap at /roadmap (under Later).
Support + ops
How do I contact support?
+
Email hello@anvat.app or message @anvat_app on Telegram — the same team operates both, replies are typically within hours.
Is there a status page?
+
Yes — /status. The widget makes a real fetch from your browser to api.anvat.app/healthz, so the signal reflects your edge specifically.
Where can I see what shipped recently?
+
The leaderboard at /leaderboard updates as models and prices change. The industry digest at /news tracks the upstream provider releases driving those changes.
Do you have a public roadmap?
+
Yes — /roadmap. Now / Next / Later with rationale for each item. If you need something soon that isn't listed, email and we'll bump priority if it's broadly useful.
How do I delete my account?
+
Email hello@anvat.app from the account address. We confirm in writing, delete within 7 business days, and refund any unused prepaid balance.
Question not here?
Email hello@anvat.app or ping @anvat_app on Telegram. If it's broadly useful, the answer ends up here.