One API for OpenAI, Anthropic and content checks

withConflux is OpenAI-compatible. Point any OpenAI SDK at our base URL, use your withConflux key, and call GPT or Claude models. The same base URL checks images, video and text for moderation, AI generation and deepfakes.

Quickstart

1. Create a free accountSign up at withconflux.com/get-started. No card needed.
2. Copy your keyDashboard › API keys. Use a cflx_test_ key while you build, a cflx_live_ key in production.
3. Make your first callUse the official OpenAI SDK, or plain HTTP.

import OpenAI from "openai";

const ai = new OpenAI({
  baseURL: "https://withconflux.com/v1",
  apiKey: process.env.CONFLUX_KEY,
});

const res = await ai.chat.completions.create({
  model: "claude-sonnet-5-5", // or "gpt-6-astra", any model in the list
  messages: [{ role: "user", content: "Hello" }],
});

console.log(res.choices[0].message.content);

Keys and authentication

Send your key as a bearer token on every request. Keys are created, paused and revoked in the dashboard; give each app or service its own key so its usage shows separately.

Authorization: Bearer cflx_live_…

Base URL

https://withconflux.com/api/v1 also works. Both reach the same API.

Chat completions

Same request and response as OpenAI's chat completions, for OpenAI and Anthropic models alike. Streaming (stream: true) and tool calls work for both. Errors come back in OpenAI's format, so SDK clients show them as usual.

const stream = await ai.chat.completions.create({
  model: "gpt-6-astra",
  messages: [{ role: "user", content: "Write a tagline" }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

Models

Returns the models on the gateway, each with the price for your account. Use the id as model. Every model and its price is also listed on withconflux.com/llm-gateway/models.

const { data } = await ai.models.list();
console.log(data.map((m) => m.id));

Embeddings

Same as OpenAI's embeddings endpoint, for example text-embedding-3-small.

const emb = await ai.embeddings.create({
  model: "text-embedding-3-small",
  input: "withConflux docs",
});

BYOK or Managed

Check an asset

Send an asset and the checks you need. Each check runs on its own and returns its own result.

curl https://withconflux.com/v1/check \
  -H "Authorization: Bearer $CONFLUX_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "asset": { "url": "https://example.com/upload.jpg" },
    "needs": ["moderation", "ai-image", "deepfake"]
  }'

Checks you can ask for

Images and videos go in asset.url (a public link to the file); text goes in asset.text.

{ "asset": { "text": "The comment to check" }, "needs": ["text-moderation"] }

{ "asset": { "url": "https://example.com/clip.mp4" },
  "needs": ["video-moderation", "ai-video"] }

Response

A valid key gets 200 back, with each check's outcome in results[] and its own status. Build on verdict and score: they have the same shape for every check.

{
  "ok": true,
  "org": "Acme Media",
  "requestId": "req_q3XbT0kM1sVz9yAa",
  "contentHash": "4f1c9e2ab07d63f58e21c4d09b7a6e35f2d8c1b0",
  "cachedCount": 1,
  "latencyMs": 412,
  "keyMode": "managed",
  "results": [
    { "need": "moderation", "status": 200, "cached": false, "mode": "live",
      "verdict": "clean", "score": 0.03, "detail": { … } },
    { "need": "ai-image", "status": 200, "cached": false, "mode": "live",
      "verdict": "ai-generated", "score": 0.94, "detail": { … } },
    { "need": "deepfake", "status": 200, "cached": true, "mode": "cache",
      "verdict": "real", "score": 0.07 }
  ]
}

A check that can't run comes back with its status and reason, while the other checks in the request still run:

{ "need": "ai-image", "status": 402, "cached": false, "mode": "none",
  "error": "Monthly spend cap reached." }

Repeat files

Every asset is recognised by a hash of what you send: the URL or the text. If you already have a content hash (for example the file's SHA-256), send it as asset.hash and the same file is recognised under any URL. If the same asset and check were seen in the last 30 days, the stored result comes back with cached: true and you are not billed again. You don't need your own cache.

Spend caps and limits

Set a monthly spend cap in the dashboard. Once it is reached, paid calls are refused with 402 and nothing is billed; repeat-file results stay free. The cap is checked as each request starts, so calls already in flight can finish a little over it.

Each key has a request limit per minute. Above it you get 429 with a Retry-After header in seconds. Each key can also be paused on its own in the dashboard.

Errors

When the whole request is refused, the status code says why. The LLM endpoints answer in OpenAI's error format ({ "error": { "message", "type", "code" } }); /v1/check answers { "ok": false, "error" }.

Per-check status in results[]

/v1/check answers 200 when the key is valid; each check then carries its own status.

Pricing

Free to start: 10,000 content checks a month, and €40 in credits for models on Managed. While you are on the free credits, the most expensive models and very long replies are held back until you add balance. BYOK is a flat €19 a month with no token markup. Full plans on withconflux.com/pricing.