Get started

Integrate Cran into your app

Your product never holds provider keys. It authenticates to Cran with a connection token (cran_…), and the dashboard controls which model, prompt, and parameters each call uses.

Native SDK (recommended)

The cran-sdk client has no runtime dependencies and returns the resolved model alongside the content. model: "auto" applies whatever you publish in the dashboard.

import { createCran } from "cran-sdk";

const { content, model } = await createCran().chat(
  [{ role: "user", content: "Summarize this ticket…" }],
  { model: "auto" } // applies your published architecture
);

OpenAI-compatible (drop-in)

Already using the OpenAI SDK? Change two values - the base URL and the key - and keep your code. The key is your Cran connection key (the optional cran link routing key, separate from the cran login MCP connection), not a real OpenAI key.

from openai import OpenAI

client = OpenAI(
    base_url="https://trycran.in/api/v1",
    api_key="cran_…",  # your Cran token, NOT an sk-… key
)
client.chat.completions.create(model="auto", messages=[...])

Embeddings & images

The same token and base URL work for the other modalities. Send model: "auto" or a catalog id (e.g. openai/text-embedding-3-small, openai/gpt-image-1):

bash

curl https://trycran.in/api/v1/embeddings \
  -H "Authorization: Bearer cran_…" -H "Content-Type: application/json" \
  -d '{ "model": "auto", "input": "text to embed" }'

Why traffic must go through Cran

Spend tracking, workflow discovery, live architecture editing, and model comparison all require requests to pass through Cran’s API. Calls that bypass Cran and hit a provider directly are invisible to the dashboard - there is no out-of-band reconciliation.

Next steps

Control plane

Publish the architecture that 'auto' resolves to.

Proxy API

Endpoints, resolution modes, and headers.

Audits

Pick the model behind 'auto' with evidence.