Decision models

Decision models classify or score text and return structured answers instead of generated chat messages. Call them through the LLM Gateway. Each model uses its own API:

Model	API	Endpoint	Credentials
TypeSafe (Jev)	System One	`/v1/systemone`	`TYPESAFE_API_KEY` provider secret
OpenAI decision models	OpenAI Decisions	`/openai/v1/decisions`	`OPENAI_API_KEY` provider secret

Both models require a workspace-scoped LangSmith API key and the matching workspace provider secret. Set your LangSmith API key before you run the examples on this page:

export LANGSMITH_API_KEY="<your-api-key>"

Use TypeSafe (Jev)

The gateway supports TypeSafe decision models with bring-your-own-key (BYOK). Configure TYPESAFE_API_KEY as a workspace provider secret. See TypeSafe setup to create a key and install the LangChain integration. TypeSafe (Jev) uses the System One API. Pass the text to evaluate in state and one or more named questions in questions. System One supports three question types:

noul: Returns the probability that the answer is true.
choice: Classifies the state into one of the supplied options.
score: Scores the state against a rubric.

The response contains answers keyed by question name rather than a chat message. Streaming is not supported. The typesafe/ prefix routes the request through your workspace’s TypeSafe provider secret. Pass your LangSmith API key, not your TypeSafe API key, as the SDK’s api_key or apiKey. Use the gateway base URL without /v1 for the SDKs:

Python
JavaScript
cURL

Install the TypeSafe SDK:

pip install typesafe-sdk

import os

from typesafe_sdk import Noul, TypeSafeClient

client = TypeSafeClient(
    api_key=os.environ["LANGSMITH_API_KEY"],
    base_url="https://gateway.smith.langchain.com",
)

response = client.system_one(
    state="Hello!",
    model="typesafe/jev-1.13.0",
    questions={
        "is_helpful": Noul(
            instructions="Does this explain what an LLM gateway does?"
        ),
    },
)

Install the TypeSafe SDK:

npm install @typesafe-ai/sdk

import { noul, TypeSafeClient } from "@typesafe-ai/sdk";

const client = new TypeSafeClient({
  apiKey: process.env.LANGSMITH_API_KEY,
  baseURL: "https://gateway.smith.langchain.com",
});

const response = await client.systemOne({
  state: "Hello!",
  model: "typesafe/jev-1.13.0",
  questions: {
    is_helpful: noul("Does this explain what an LLM gateway does?"),
  },
});

curl https://gateway.smith.langchain.com/v1/systemone \
    -H "Authorization: Bearer $LANGSMITH_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "state": "Hello!",
      "model": "typesafe/jev-1.13.0",
      "questions": {
        "is_helpful": {
          "type": "noul",
          "instructions": "Does this explain what an LLM gateway does?"
        }
      }
    }'

Use OpenAI decision models

OpenAI decision models use the OpenAI Decisions API, which has its own request and response format, separate from System One. The gateway supports the OpenAI Decisions API through direct model access. Configure OPENAI_API_KEY as a workspace provider secret. Send requests to POST /openai/v1/decisions using your workspace-scoped LangSmith API key. Set model to the native OpenAI model name without the openai/ prefix:

curl https://gateway.smith.langchain.com/openai/v1/decisions \
    -H "Authorization: Bearer $LANGSMITH_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "gpt-6-luna",
      "input": "The package arrived with a broken screen.",
      "questions": [
        {
          "type": "predicate",
          "name": "damaged",
          "instructions": "Does the customer report a damaged item?"
        }
      ]
    }'

The gateway preserves the native request and response format:

{
  "model": "gpt-6-luna",
  "answers": [
    {
      "type": "predicate",
      "name": "damaged",
      "probability": 1.0
    }
  ],
  "usage": {
    "input_tokens": 164,
    "input_tokens_details": {
      "cached_tokens": 0,
      "cache_write_tokens": 0
    },
    "output_tokens": 0,
    "output_tokens_details": {
      "reasoning_tokens": 0
    },
    "total_tokens": 164
  }
}

Gateway access, rate-limit, and budget policies apply to Decisions requests, and token usage counts toward spend caps. The gateway traces each request to LangSmith with the request as input and the answers and token usage as output.

Get started

Core capabilities

Administration and governance

Advanced

Use TypeSafe (Jev)

Use OpenAI decision models

See also

​Use TypeSafe (Jev)

​Use OpenAI decision models

​See also

Use TypeSafe (Jev)

Use OpenAI decision models

See also