Question 1

What is DeFAI Relay?

Accepted Answer

A gateway, compatible with both OpenAI and Anthropic, that puts one endpoint in front of every frontier model. You sign in with your wallet, top up in USDC on Base, and pay per token — metered and settled on-chain. No email accounts, no cards, no monthly plans.

Question 2

Is it really OpenAI-compatible?

Accepted Answer

Yes. Point any OpenAI SDK at https://defai.gtio.work/v1 with a relay key. /chat/completions, /responses, /completions, /embeddings, /images/generations, and /models all work — including streaming and tool calls. It's usually a two-line change. See the docs.

Question 3

Can I use Claude Code, Codex, Cline or Cursor?

Accepted Answer

Yes. The relay speaks both the OpenAI and Anthropic wire formats. Claude Code points ANTHROPIC_BASE_URL at https://defai.gtio.work (it adds /v1/messages itself) with ANTHROPIC_AUTH_TOKEN set to your relay key. Codex, OpenClaw, Hermes, Kilo Code, OpenHands, Qwen Code, pi, Cline, Roo Code, Zed and Cursor use the OpenAI-compatible base https://defai.gtio.work/v1. Copy-paste setups for each are on the integrations page.

Question 4

Can I generate images?

Accepted Answer

Yes — three ways. gpt-image runs on the OpenAI Images API (POST /v1/images/generations). Imagen and Gemini image models (Nano Banana, Qwen-Image, and more) run on the Google GenAI / Vertex surface at /vertex-ai. Both image-to-image (editing / reference images) and text-to-image work. Images bill per generation or per output token depending on the model; the per-model rate is on the pricing page. There is also an image mode in the in-console Playground.

Question 5

Can I generate video?

Accepted Answer

Yes. Veo, Seedance and other video models run on the Google GenAI / Vertex surface (/vertex-ai) as an async job — submit a generation, then poll the operation until the video is ready. Video bills per second of output, by resolution. Point the google-genai SDK (Vertex mode) at the relay; setup is on the integrations page.

Question 6

Do you support prompt caching?

Accepted Answer

Yes, and it lowers your bill. Cached input tokens are billed at the model's cache-read rate — typically a fraction of the fresh-input price — and cache writes at their own rate, normalized across the OpenAI and Anthropic usage formats. You pay what caching actually costs, not full price for cached tokens.

Question 7

How do I pay?

Accepted Answer

Top up your balance in native USDC on Base. Every call is metered by prompt + completion tokens and debited from your prepaid balance in integer USDC base units. There's no subscription — you pay only for the tokens you use. Live rates are on the pricing page.

Question 8

Do I need gas or a credit card?

Accepted Answer

No card, ever. The only on-chain action is your USDC deposit transaction, which costs a few cents of gas on Base. Signing in with your wallet (SIWE) is a plain signature — gasless.

Question 9

Which models can I call?

Accepted Answer

120+ models across OpenAI, Anthropic, Google, DeepSeek, Qwen, xAI and more — all behind one endpoint. Browse the full list with live prices on the pricing page. New models appear automatically as the upstream lists them.

Question 10

How is pricing calculated?

Accepted Answer

Per token. Each model's input/output price comes from the live upstream catalog; we apply a single margin multiplier and charge ceil((prompt×in + completion×out) × markup) in integer USDC base units (6 decimals). No rounding in anyone's favor.

Question 11

What if a model has no published price?

Accepted Answer

It's refused, not billed at a guess. A request to an unpriced model returns an error instead of charging an estimate. We never settle money against a price we can't verify.

Question 12

Do you store my prompts?

Accepted Answer

No. We record only what's needed to bill you — model and prompt/completion token counts and cost. Prompt and completion content is proxied straight to the provider and not persisted by us.

Question 13

How fast are deposits credited?

Accepted Answer

Your USDC transfer to the treasury is picked up after a handful of Base confirmations (seconds to about half a minute) and credited to the balance of the wallet that sent it — the same wallet you sign in with.

Question 14

Is streaming supported?

Accepted Answer

Yes. Set "stream": true and Server-Sent Events pass straight through with backpressure — tokens arrive the instant the model emits them, with no buffering at the edge.

Question 15

How do API keys work?

Accepted Answer

Mint relay keys (sk-relay-…) in the console; the plaintext is shown exactly once, and only its hash is stored. Revoke any key instantly. Keys authenticate the relay; your wallet session is only for the console.

Question 16

How do I reach support?

Accepted Answer

Email gtoxlili@outlook.com — bug reports, billing questions, model requests, anything. Include your wallet address (never your key) and a request_id from the error response if you have one, and we can trace the exact call.

Question 17

Is my balance handled safely?

Accepted Answer

Balances are integer USDC base units backed by on-chain deposits, with an up-front spend hold reserved per request and reconciled to actual usage. Settlement is deliberate and verifiable — money is treated as money.

Questions, answered.

What is DeFAI Relay?+

Is it really OpenAI-compatible?+

Can I use Claude Code, Codex, Cline or Cursor?+

Can I generate images or video?+

Do you support prompt caching?+

How do I pay?+

Do I need gas or a credit card?+

Which models can I call?+

How is pricing calculated?+

What if a model has no published price?+

Do you store my prompts?+

How fast are deposits credited?+

Is streaming supported?+

How do API keys work?+

How do I reach support?+

Is my balance handled safely?+