API reference

Kymo exposes an OpenAI-compatible chat completions API. Any OpenAI or Groq SDK works by changing the base URL to /api/public/v1.

Quickstart

Create an API key on the API keys page, then call the endpoint:

bash

curl https://your-kymo-domain/api/public/v1/chat/completions \
  -H "Authorization: Bearer $KYMO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama-3.3-70b-versatile",
    "messages": [
      { "role": "system", "content": "You are a concise assistant." },
      { "role": "user", "content": "What is an LPU?" }
    ],
    "temperature": 0.7
  }'

Or with the OpenAI Python SDK:

python

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["KYMO_API_KEY"],
    base_url="https://your-kymo-domain/api/public/v1",
)

completion = client.chat.completions.create(
    model="openai/gpt-oss-120b",
    messages=[{"role": "user", "content": "Write a haiku about latency."}],
)
print(completion.choices[0].message.content)

Authentication

Send your key as a bearer token: Authorization: Bearer kymo_…. Keys are stored only as SHA-256 hashes, so the full value is shown exactly once at creation. Revoked keys immediately return 401.

Chat completions

POST /api/public/v1/chat/completions

model — required, see Models.
messages — required array of system/user/assistant messages.
temperature, top_p, max_tokens, stop — optional sampling controls.
stream — boolean, emits server-sent events.
response_format — { "type": "json_object" } for JSON mode.

GET /api/public/v1/models lists the available models in OpenAI format.

Streaming

Set stream: true to receive incremental deltas as SSE. The stream ends with data: [DONE].

typescript

const res = await fetch(BASE_URL + "/chat/completions", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.KYMO_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "llama-3.3-70b-versatile",
    messages: [{ role: "user", content: "Count to five." }],
    stream: true,
  }),
});

// Server-sent events: data: {...}\n\n ... terminated by data: [DONE]
const reader = res.body!.getReader();

JSON mode

JSON mode guarantees syntactically valid JSON. Describe the exact shape you want in a system message, and avoid combining it with streaming.

json

{
  "model": "llama-3.3-70b-versatile",
  "response_format": { "type": "json_object" },
  "messages": [
    { "role": "system", "content": "Reply with JSON: { title, tags[] }" },
    { "role": "user", "content": "Summarise the article about GPUs." }
  ]
}

Vision

Vision-capable models accept an array content block with image URLs or base64 data URLs. Up to 5 images per request, 20 MB each.

json

{
  "model": "qwen/qwen3.6-27b",
  "messages": [{
    "role": "user",
    "content": [
      { "type": "text", "text": "What is in this image?" },
      { "type": "image_url", "image_url": { "url": "https://example.com/photo.jpg" } }
    ]
  }]
}

Models

llama-3.3-70b-versatile

General-purpose workhorse. Fast, strong at reasoning and long-form text.

128K ctx

JSON

openai/gpt-oss-120b

Large open-weight model for demanding reasoning and code tasks.

128K ctx

JSON

qwen/qwen3.6-27b

Multimodal: image understanding, OCR, captioning, visual Q&A.

131K ctx

Vision

JSON

groq/compound

Agentic system with built-in web search, page visits, and code execution.

128K ctx

groq/compound-mini

Single tool call per request, ~3x lower latency than Compound.

128K ctx

Rate limits & errors

Each API key is limited to 60 requests per rolling 60-second window. Exceeding it returns 429.

400 — invalid body or unknown model.
401 — missing, invalid or revoked key.
429 — rate limit exceeded.
502 — upstream inference error.

Every request is recorded on your usage dashboard, including failures.