WebFV
AI API gateways / Field guide

CodeCraft API: endpoint setup, pricing and billing caveats

Identify the CodeCraft API gateway, read its documented client settings and distinguish advertised plans from verified billing, model availability and refund guarantees.

By WebFV Editorial · Checked October 28, 2026 · Public documentation snapshot · October 28, 2026

What the evidence coversProvider website, quickstart, pricing, model documentation, terms, privacy and refund policy reviewed. No account registration, purchase, deposit, key creation or API/model request performed.

CodeCraft API at codecraftapi.com describes itself as a third-party AI API gateway. Its public documentation specifies an OpenAI-compatible interface, monthly token packages and a separate prepaid balance. This guide concerns that exact domain, not another project called CodeCraft or an API sold directly by a model developer.

We reviewed its public pages on October 28, 2026. We did not register, deposit funds, subscribe, generate credentials or send an API/model request. Compatibility, model routing and billing behavior below are the service's documented claims, not results from our own integration test.

Identify the service before connecting

The Terms of Service, effective August 23, 2026, identify the operator as Radox AI. They say the gateway routes requests to third-party providers and does not build or host the underlying models. They also state that platform model names may differ from upstream names.

That distinction matters when choosing a model. A familiar name in a gateway catalog does not independently verify the upstream model version, provider relationship or output quality. We have not established an authorized partnership with any model developer, or verified the provenance of individual routes. Use the official website and its own documentation to identify the account and endpoint you intend to use.

Documented endpoint and client settings

The Getting Started guide gives these settings for tools that accept a custom OpenAI-compatible provider:

Setting Documented value
Provider type OpenAI Compatible
Base URL on the apex-domain guide https://codecraftapi.com/v1
API key A CodeCraft-issued key, described as beginning with cc_
Model An available model ID from its model list
Chat route /chat/completions under that base URL

The documented key-creation path is Dashboard > API Keys > Create Key. Its request examples use Bearer authentication. These are instructions for readers who choose to create an account; no key was created for this article.

The www homepage instead shows https://www.codecraftapi.com/v1 and an sk-... placeholder in its example. We have not tested whether authenticated calls behave identically across those hosts. Follow the documentation associated with your account and ask support to resolve a mismatch before sending a real key; do not assume an HTTP redirect preserves authentication correctly.

“OpenAI-compatible” here is a claim about an API interface. It is not evidence of OpenAI affiliation or a guarantee that every SDK method, coding-agent version, streaming path or tool-call behavior works unchanged. This guide therefore provides no copied inference command presented as a tested example.

Published package prices and the free-tier discrepancy

The pricing page listed these monthly USD packages when checked:

Package Monthly price Advertised monthly tokens
Starter $1.20 30M
Basic $3.50 100M
Pro $5.75 200M
Scale $11.50 500M
Unlimited $50 No monthly token cap

These are the provider's advertised amounts, not a verified checkout total or a comparison with buying directly from upstream vendors. The table uses monthly billing, rather than the lower monthly equivalent shown beside annual billing.

The free allowance was inconsistent: the apex homepage and pricing page showed 1M tokens per month, while the www homepage showed 0.5M. Treat the available free quota as unresolved until the account's current plan confirms it. No promotional code or extra free-credit offer was verified.

The pricing page says input and output both count, and the Unlimited package retains per-minute limits: 3,000 requests and 5M tokens per minute. A package called Unlimited should not be read as a promise of unrestricted throughput.

Allowance, balance and model rates are different

The terms describe monthly allowance being spent first, with unused allowance expiring at the billing reset. After it runs out, requests draw from prepaid balance at the selected model's published token rate. The service says requests return HTTP 402 when both sources are exhausted. We did not verify that enforcement or metering.

The model API documentation defines model IDs, context windows, capability flags and separate input/output prices. Its pricing.input_per_1k and pricing.output_per_1k fields use USD per 1,000 tokens. Convert units before comparing those values with prices displayed per million tokens.

Check model-specific capabilities before assuming that image input, tools or JSON output are available. The documentation says only active public models are listed and an unavailable model can return 404. A static example ID is not proof that the route remains active. We have not queried the authenticated catalog and do not claim that every advertised model is currently callable.

Cancellation and refunds: resolve the policy mismatch

We did not find a public, verified click path for cancelling subscription renewal. The terms describe closing an account in settings; account closure and stopping a future renewal should not be treated as interchangeable without confirmation. Before paying, obtain the renewal terms and cancellation method for the chosen payment route. Keep the order reference and confirmation of any cancellation.

The Refund Policy advertises a seven-calendar-day guarantee for a first subscription, excluding ordinary renewals and plan changes. It separately allows requests for unused top-up balance within 14 days, subject to account status and processor fees. Consumed tokens are excluded. Crypto payments have additional restrictions, including no fiat refund.

However, the terms say plan fees are refundable pro rata only for a material service failure that cannot be remedied. These statements do not fully align. Ask the operator which policy governs your transaction rather than assuming a refund is automatic. The refund page asks for the account email, order or transaction reference, and reason through its listed support contact.

Data handling before sending private code

The Privacy Policy says prompt and response bodies are forwarded rather than stored in its database, but usage metadata is retained. It also allows truncated error logs that can contain request fragments. Upstream providers receive the requests and apply their own retention policies.

Those are policy statements, not an independent audit or an end-to-end zero-retention guarantee. For confidential code, identify the actual upstream route and applicable policy before use. Keep credentials outside client-side code and public repositories, and use only the permissions needed for the intended integration.

If your decision depends on an exact free quota, specific upstream identity, automatic-renewal terms or a guaranteed refund, the public material reviewed here leaves gaps. Resolve those points with the service before paying or moving an existing workload.