Your content, your AI,
anywhere you need it.

1 Get your API key

An API key maps to exactly one deployment — one chatbot trained on one set of documents. The key alone decides which documents an answer can come from, so you never pass a deployment id separately.

1. Create a deployment and train it In your dashboard go to Document Training, create a deployment, upload your files (PDF, DOCX, TXT, CSV) and click Train. Wait for the status to reach Completed.
2. Open the API key page Go to API key in the sidebar, or click here if you're signed in.
3. Generate a key for that deployment Each deployment gets its own key. Copy it immediately — it is stored hashed, so it cannot be shown again. If you lose it, delete the key and generate a new one.
4. Keep it server-side where you can A key spends your token balance. Our chat widget uses it in the browser by necessity — if you do the same, restrict the deployment by IP in Settings → IP Restriction and watch usage under Reports.

2 Authentication

Send your key whichever way suits the client. All three are equivalent:

Authorization: Bearer YOUR_API_KEY     # preferred, used by /api/v1/*
X-API-KEY: YOUR_API_KEY                # header alternative
?api_key=YOUR_API_KEY                  # query/form field — required for SSE,
                                       # because EventSource cannot set headers

Keys are stored as SHA-256 hashes. An invalid or disabled key, or a deployment that is not published, returns 401.

3 Quickstart

Ask your first question in one request. Pick a language:

curl https://askcybex.com/api/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "YOUR_DEPLOYMENT_ID",
    "messages": [
      { "role": "user", "content": "What is your refund policy?" }
    ]
  }'

// Works with the official OpenAI SDK — only the baseURL changes.
import OpenAI from "openai";

const client = new OpenAI({
  apiKey:  "YOUR_API_KEY",
  baseURL: "https://askcybex.com/api/v1",
});

const res = await client.chat.completions.create({
  model: "YOUR_DEPLOYMENT_ID",
  messages: [{ role: "user", content: "What is your refund policy?" }],
});

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

# pip install openai
from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://askcybex.com/api/v1",
)

res = client.chat.completions.create(
    model="YOUR_DEPLOYMENT_ID",
    messages=[{"role": "user", "content": "What is your refund policy?"}],
)

print(res.choices[0].message.content)

$ch = curl_init('https://askcybex.com/api/v1/chat/completions');
curl_setopt_array($ch, [
    CURLOPT_POST           => true,
    CURLOPT_RETURNTRANSFER => true,
    CURLOPT_HTTPHEADER     => [
        'Authorization: Bearer YOUR_API_KEY',
        'Content-Type: application/json',
    ],
    CURLOPT_POSTFIELDS     => json_encode([
        'messages'        => [
            ['role' => 'user', 'content' => 'What is your refund policy?'],
        ],
        'conversation_id' => 'visitor-123',
    ]),
]);
$data = json_decode(curl_exec($ch), true);
curl_close($ch);

echo $data['choices'][0]['message']['content']
     ?? $data['error']['message'];

Multi-turn conversations

Follow-up questions need the earlier turns. Pick one of two styles — don't mix them.

Style A — resend the messages (OpenAI-native). Stateless, and what the OpenAI SDKs and Open WebUI do automatically:

{
  "model": "98",
  "messages": [
    { "role": "user",      "content": "What is Hogwarts?" },
    { "role": "assistant", "content": "Hogwarts is a school of witchcraft and wizardry..." },
    { "role": "user",      "content": "Who are the students there?" }
  ]
}

Style B — reuse the conversation id. Send only the new question; we rebuild the history server-side:

# Turn 1 — no id sent, so one is minted and returned
curl -X POST https://askcybex.com/api/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"98","messages":[{"role":"user","content":"What is Hogwarts?"}]}'

# response contains BOTH ids — copy the conv- one:
#   "id":              "chatcmpl-f616e6134ae341316ba712a9"   <- NOT this
#   "conversation_id": "conv-dc4d05c139d35b910ff4be69"       <- this one

# Turn 2 — reuse conversation_id, send only the new question
curl -X POST https://askcybex.com/api/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"98",
       "messages":[{"role":"user","content":"Who are the students there?"}],
       "conversation_id":"conv-dc4d05c139d35b910ff4be69"}'

If you send both, the messages history wins, so a turn is never counted twice.

4 Endpoint reference

One endpoint answers questions. The rest are the things a chat widget needs around a conversation, and one for teaching a chatbot something new. Every one of them takes the same bearer key, and the key is what selects the chatbot — there is no deployment id to pass.

Chat

const es = new EventSource(
  "https://askcybex.com/api/v1/chat/completions?stream=1" +
  "&api_key=" + YOUR_API_KEY +
  "&message=" + encodeURIComponent(question) +
  "&conversation_id=" + conversationId
);

es.onmessage = (e) => {
  if (e.data === "[DONE]") return;
  const chunk  = JSON.parse(e.data);
  const choice = chunk.choices[0];

  if (choice.delta.content) output.textContent += choice.delta.content;

  if (choice.finish_reason) {
    es.close();
    rate(chunk.chat_log_id);        // /widget/feedback
    remember(chunk.conversation_id);
  }
};

Widget

Documents

5 Use it inside tools you already run

Because the API speaks the OpenAI format, most AI tooling connects without any code. For Open WebUI:

1. Settings → Connections → OpenAI APIAdd a new connection.
2. Base URLhttps://askcybex.com/api/v1
3. API keyYour deployment key.
4. Save, then pick your deployment from the model listIt appears by deployment id.

The same pattern works for LangChain, LlamaIndex, agent frameworks, ERP and CRM add-ons, and any internal tool that lets you set an OpenAI base URL.

6 Rate limits, errors & billing

Every response carries X-RateLimit-Limit and X-RateLimit-Remaining. Exceeding a limit returns 429.

Billing. Answers consume tokens from your account balance. When a question cannot be grounded in your documents, the refusal is returned before the language model is called — so a “Sorry, I don’t know” costs you nothing.