The same Outcry, now an endpoint.

curl http://localhost:11434/v1/chat/completions \
  -H "Authorization: Bearer ocry_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "outcry",
    "messages": [
      {"role": "user", "content": "Draft an opening line for a press release about a wage theft case."}
    ],
    "stream": true
  }'

OpenAI /v1/chat/completions shape — streaming, non-streaming, multi-turn, system prompts, stop sequences, seed. Tool calling and structured JSON output aren't supported yet.

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ocry_...",  # issue per-tool in the Keys tab
)

resp = client.chat.completions.create(
    model="outcry",
    messages=[
        {"role": "system", "content": "You draft union comms."},
        {"role": "user", "content": "30-word announcement about Thursday's rally."},
    ],
    stream=True,
)

for chunk in resp:
    print(chunk.choices[0].delta.content or "", end="")

Same for the Node SDK, the Go SDK, LangChain, LlamaIndex, instructor, anything that takes a base URL.