How to Use the Radox AI API in 2026: From Sign-Up to a Streaming Reply in Five Steps
If your code already talks to an OpenAI-style API, connecting to Radox AI is a short job: an account, a key, a base URL and a model name. This guide walks through each step with copy-ready examples (placeholder keys only), then covers streaming, rate limits, how tokens are counted and what HTTP 402 means. This article focuses on chat/text.
| Base URL | radoxai.com/v1 |
| Auth | Bearer token (your key) |
| Format | OpenAI-compatible |
| Streaming | Server-sent events |
| Out of funds | HTTP 402 |
| Cost to start | Free, no card |
Five small steps, and only one of them is new.
The new part is the base URL. The key, the request shape and the streaming loop are the ones you already know.
Radox's own pitch is "Swap your base URL. Keep your code." and its homepage says it is "Ready in about a minute". The steps below follow the same order as the three-step flow on the homepage (create a key, point your client, ship), with the detail a first-time user usually wants spelled out.
From nothing to a first reply.
Sign up on radoxai.com. Sign-in with Google or GitHub is offered.
Generate it in the dashboard and scope it to what you need.
https://radoxai.com/v1
Use a model ID such as claude-sonnet-5.
Same request shape as any OpenAI-style client.
The dashboard tracks requests, tokens and spend.
Create the account and read three rules first.
The Terms set a few conditions at sign-up: "You must be at least 16, or the age of digital consent in your country, whichever is higher." and "One person or organisation per account." They add that "Creating multiple accounts to obtain extra free allowances is a breach of these terms."
The Privacy Policy says that if you "sign in with Google or GitHub", Radox receives your name, email address and profile picture URL from that provider, and not your password for those services.
Create the key and scope it.
The homepage describes the step as: "Sign up and generate an API key from your dashboard. Scope it to inference, models, or embeddings." The docs index lists "Bearer token authentication, key scopes, and rate limits" under Authentication.
sk-xxxxxxxx, a placeholder.Base URL, model ID and a first reply in three languages.
Set the base URL to https://radoxai.com/v1, use your key, and name a model. The model IDs are the lowercase strings in the catalog on radoxai.com/models.
About the key in these examples: a real key has the prefix shown in your own dashboard (the Radox Getting Started page shows keys beginning cc_ followed by 48 characters). sk-xxxxxxxx is only a placeholder, not a real format.
The compatibility claim behind this is on the FAQ: "Request and response shapes match the OpenAI spec, including streaming chunks, tool calls, and error envelopes." The official SDKs for Python, Node, Go and .NET are the documented route.
| Method | Endpoint | Purpose |
|---|---|---|
| GET | /models | List available models |
| GET | /models/{id} | Retrieve one model |
| POST | /chat/completions | Generate a chat completion |
| POST | /embeddings | Create embeddings |
| Field | Value | Where it goes |
|---|---|---|
| Base URL | https://radoxai.com/v1 | Client setting |
| Key | Your key (placeholder sk-xxxxxxxx here) | Authorization: Bearer header |
| model | A lowercase ID from radoxai.com/models | JSON body |
| messages | List of role and content objects | JSON body |
| Setting | Value |
|---|---|
| API Provider | OpenAI Compatible |
| Base URL | https://radoxai.com/v1 |
| API Key | Your Radox key |
| Model | Any model ID from /models |
| Language | Route |
|---|---|
| Python | Official OpenAI SDK |
| Node.js | Official OpenAI SDK |
| Go | Official OpenAI SDK |
| .NET | Official OpenAI SDK |
| Anything else | Any tool that accepts a custom endpoint |
| Setting | Python SDK | Node.js SDK | cURL |
|---|---|---|---|
| Base URL | base_url | baseURL | Start of the URL |
| Key | api_key | apiKey | -H "Authorization: Bearer ..." |
| Streaming flag | stream=True | stream: true | "stream": true in the JSON body |
Add stream=True and read the chunks.
Radox lists "Server-sent events with keep-alive pings and hour-long connections". In practice that means the same streaming loop you would write for any OpenAI-style client:
For reasoning models, the homepage says the chain-of-thought output "arrives in a separate field", so the visible answer stays clean. The models page labels which models support tool calling and vision input; for the request format, check the API reference (JAVIS inference, not in the Radox docs: that they share the same endpoint).
JAVIS diagram from the homepage, the pricing page and the Terms, Oct 15, 2026. Step 3 is JAVIS wording for model routing.
How tokens are counted, and what the request path looks like.
Tokens count prompt and completion combined. Each plan has a monthly allowance that is spent first; after it, requests bill from your account balance at the model's rate; when both are gone, the request is refused. The Terms word the last step this way: "requests are refused with HTTP 402 before reaching a model".
Where an upstream provider does not report exact token counts, the Terms say "we estimate them using a documented method". Plan prices and per-model overage rates are covered in the plan and model guides; this page only needs the order above.
Per-minute ceilings by plan, and where to read them.
The pricing page states: "Rate limits are enforced per minute and reported in X-RateLimit-* response headers." Read those headers in your client when you plan batch jobs.
| Plan | Requests / min | Tokens / min |
|---|---|---|
| Free | 60 | 100,000 |
| Starter | 120 | 200,000 |
| Basic | 300 | 500,000 |
| Pro | 600 | 1,000,000 |
| Scale | 1,200 | 2,000,000 |
| Unlimited | 3,000 | 5,000,000 |
| Tokens / request | Free | Pro | Unlimited |
|---|---|---|---|
| 500 | 60 req cap | 600 req cap | 3,000 req cap |
| 1,000 | 60 req cap | 600 req cap | 3,000 req cap |
| 2,000 | 50 token cap | 500 token cap | 2,500 token cap |
| 5,000 | 20 token cap | 200 token cap | 1,000 token cap |
| 10,000 | 10 token cap | 100 token cap | 500 token cap |
| Plan | Price / month | Tokens / month | Per 1M tokens | Requests / min |
|---|---|---|---|---|
| Free | Free | 10M | not listed | 60 |
| Starter | $4.50 | 100M | $0.045 | 120 |
| Basic | $8.50 | 200M | $0.0425 | 300 |
| Pro | $12.50 | 350M | $0.0357 | 600 |
| Scale | $15.50 | 500M | $0.031 | 1,200 |
| Unlimited | $50 | No cap | not listed | 3,000 |
| Plan | Tokens / month | About per day |
|---|---|---|
| Free | 10M | 0.3M |
| Starter | 100M | 3.3M |
| Basic | 200M | 6.7M |
| Pro | 350M | 11.7M |
| Scale | 500M | 16.7M |
| Plan | Tokens / min | Hours to exhaust |
|---|---|---|
| Free | 100k | 1.7 h |
| Starter | 200k | 8.3 h |
| Basic | 500k | 6.7 h |
| Pro | 1M | 5.8 h |
| Scale | 2M | 4.2 h |
| Change | Timing | Effect on allowance |
|---|---|---|
| Upgrade | Immediately | Allowance is refreshed |
| Downgrade | From the next billing period | You keep what you already paid for |
A short checklist when a first request does not work.
| Symptom | Check | Basis |
|---|---|---|
| Request refused with HTTP 402 | Allowance and balance are both exhausted: top up or upgrade | Terms section 4 and pricing FAQ |
| Requests are throttled | Read the X-RateLimit-* headers and compare with your plan's per-minute limits | Pricing page |
| Client cannot find the endpoint | Base URL must be https://radoxai.com/v1 | Homepage quickstart |
| Model not accepted | Use the exact lowercase ID from the catalog (JAVIS inference, not in the Radox docs: the error behaviour itself) | Models page |
| Authentication fails | Key sent as a Bearer token in the Authorization header | Docs index and homepage |
| Authentication still fails | Key copied in full and its scope matches the call you make | JAVIS inference, not in the Radox docs |
| What you see | Likely cause | What to do | Source |
|---|---|---|---|
| HTTP 402 | Plan allowance and account balance are both used up | Top up or upgrade; requests resume immediately | Pricing FAQ, Terms |
| X-RateLimit-* headers | Per-minute limits for your plan are reported here | Read them in the client and pace batch jobs to your plan | Pricing page |
| Check | Expected |
|---|---|
| Base URL | https://radoxai.com/v1 |
| Header | Authorization: Bearer plus your key |
| Model | Lowercase ID copied from the catalog |
| Key scope | Matches the call you make (JAVIS inference, not in the Radox docs) |
| Balance | Allowance or balance left, otherwise 402 |
Habits that protect your balance.
- "Keep keys server-side."
- "Scope each key to only what it needs, and rotate keys you suspect are exposed."
- "Usage incurred by a leaked key is your responsibility."
- One key per app or environment, so one leak is easy to revoke.
- Never paste keys or secrets into prompts.
- Check the usage dashboard after your first day of real traffic.
- Server-side environment variable
- A secrets manager
- One key per app or environment
- Browser JavaScript
- Mobile apps
- Public repositories
How this guide was made.
Links labelled as Radox AI calls to action on this page use JAVIS's referral link; if you sign up through them, JAVIS may receive a referral reward at no extra cost to you.
JAVIS built this guide from Radox AI's official pages and has not run these calls hands-on.
This guide is built from Radox AI's official pages (home, pricing, docs, models, Terms, Privacy), dated Oct 15, 2026. This article focuses on chat/text.
Start on the free plan; no card required per the pricing page.
Create a free accountRead the related guides below first.
Choose the next articleIf you are about to wire up an agent or app.
Questions people ask while connecting
What is the Radox AI base URL?
https://radoxai.com/v1. Set it as the base URL of an OpenAI-compatible client.
Do I need a card to start?
No. The pricing page states the free plan needs no card.
Which SDKs can I use?
The official OpenAI SDKs for Python, Node, Go and .NET, or any tool that accepts a custom endpoint.
How do I switch models?
Change the model string; Radox describes switching as a one-string change.
What does HTTP 402 mean here?
Both your plan allowance and your balance are exhausted, so the request is refused before it reaches a model.
Where do I see my usage?
In the dashboard, which tracks requests, tokens and spend per key and per model.
Is it safe to put my key in front-end code?
No. The Terms say to keep keys server-side and not to embed them in browser JavaScript, mobile apps or public repositories.
Where this came from, and the date it applies to.
Quickstart, three-step flow, streaming, compatibility statement: radoxai.com (home), radoxai.com/docs and radoxai.com/docs/getting-started, Oct 15, 2026.
Rate limits, billing order, 402: radoxai.com/pricing, Oct 15, 2026.
Account rules, keys, token estimates: Radox Terms of Service, effective 23 August 2026.
Sign-in options: Radox Privacy Policy, effective 23 August 2026.
Model IDs: radoxai.com/models, Oct 15, 2026.
Images: logo hosted on radoxai.com (branding).
