# Jev / TypeSafe context

Jev is a TypeSafe System One decision model. It accepts natural-language or structured state and returns typed decisions with calibrated probabilities instead of generated text.

## Primary implementation

- LangChain.js TypeSafe provider:
  https://github.com/langchain-ai/langchainjs/tree/main/libs/providers/langchain-typesafe

  Contains the JavaScript/TypeScript implementation of the `@langchain/typesafe` package, including `TypeSafeClassifier`.

## LangChain documentation

- Python TypeSafe integration:
  https://github.com/langchain-ai/docs/blob/main/src/oss/python/integrations/providers/typesafe.mdx

  Contains Python installation, API-key setup, `TypeSafeClassifier` usage, decision primitives such as `Noul`, `Choice`, and `Score`, supported state formats, LangChain message support, tracing, and agent middleware examples.

- JavaScript TypeSafe integration:
  https://github.com/langchain-ai/docs/blob/main/src/oss/javascript/integrations/providers/typesafe.mdx

  Contains JavaScript and TypeScript installation, server-side API-key setup, `TypeSafeClassifier` usage, question configuration, response accessors, supported state formats, and usage details.

- LangSmith Gateway decision models:
  https://github.com/langchain-ai/docs/blob/main/src/langsmith/llm-gateway-decision-models.mdx

  Contains instructions for calling TypeSafe/Jev through the LangSmith LLM Gateway, BYOK configuration, gateway authentication, model names such as `typesafe/jev-1.13.0`, and Python, JavaScript, and cURL examples.

## Documentation source directories

- Python provider documentation directory:
  https://github.com/langchain-ai/docs/tree/main/src/oss/python/integrations/providers

  Contains provider-specific Python integration documentation, including `typesafe.mdx`.

- JavaScript provider documentation directory:
  https://github.com/langchain-ai/docs/tree/main/src/oss/javascript/integrations/providers

  Contains provider-specific JavaScript and TypeScript integration documentation, including `typesafe.mdx`.

- LangSmith documentation directory:
  https://github.com/langchain-ai/docs/tree/main/src/langsmith

  Contains LangSmith product documentation, including LLM Gateway documentation.

## Provider index pages

- Python provider index:
  https://github.com/langchain-ai/docs/blob/main/src/oss/python/integrations/providers/all_providers.mdx

  Lists TypeSafe as a Python classification-model provider and links to its integration documentation.

- JavaScript provider index:
  https://github.com/langchain-ai/docs/blob/main/src/oss/javascript/integrations/providers/all_providers.mdx

  Lists TypeSafe as a JavaScript classification-model provider and links to its integration documentation.

## Official TypeSafe documentation

- TypeSafe models:
  https://docs.typesafe.ai/models

  Contains the official model catalog, including Jev.

- System One:
  https://docs.typesafe.ai/concepts/system-one

  Explains the System One model concept and how it differs from generative language models.

- Decision primitives:
  https://docs.typesafe.ai/primitives

  Documents the typed decision primitives used by TypeSafe questions, including Noul, Choice, and Score.

- TypeSafe API-key management:
  https://console.typesafe.ai/settings/keys

  TypeSafe console page for creating and managing API keys.

## Packages

- Python package:
  `langchain-typesafe`

  Install with:
  `pip install langchain-typesafe`

- JavaScript/TypeScript package:
  `@langchain/typesafe`

  Install with:
  `npm install @langchain/typesafe @langchain/core`

## Environment variables

- Direct TypeSafe API access:
  `TYPESAFE_API_KEY`

- Optional TypeSafe-compatible endpoint:
  `TYPESAFE_BASE_URL`

- LangSmith Gateway authentication:
  `LANGSMITH_API_KEY`

## Agent instructions

- For documentation or usage questions, read the language-specific TypeSafe integration guide first.
- For JavaScript or TypeScript implementation changes, inspect:
  https://github.com/langchain-ai/langchainjs/tree/main/libs/providers/langchain-typesafe
- For Python implementation changes, search the `langchain-ai/langchain` repository for `langchain-typesafe` and `TypeSafeClassifier`:
  https://github.com/langchain-ai/langchain
- Treat Jev as a typed decision/classification model, not as a conventional chat model.
- Jev returns structured decisions and probabilities.
- Keep `TYPESAFE_API_KEY` on the server and never expose it in browser code.
- When working with the LangSmith Gateway, distinguish `LANGSMITH_API_KEY` from the provider secret `TYPESAFE_API_KEY`.