Jev Integrations

Jev exposes one HTTP endpoint, official SDKs, and ecosystem integrations. This page is the map. Read Getting Started for installation and a first call.

The HTTP API

Everything else wraps this:

POST https://api.typesafe.ai/v1/systemone
Authorization: Bearer <TYPESAFE_API_KEY>
GET  https://api.typesafe.ai/v1/models

The request carries a model, a state, and questions. The response carries one answer per question, the versioned model ID that answered, and usage. Keep the key server-side; the SDKs intentionally block browser use.

Official SDKs

A Swift client and a Go MCP server exist in the community ecosystem, but treat unofficial clients as unverified until you have read their source.

Agent Skill

TypeSafe publishes an agent skill that documents the API contract for coding agents:

npx skills add typesafe-ai/skills --skill typesafe-ai

The Agent Skills section covers the SKILL.md standard. For agent integrations specifically, remember the routing advice from the skill routing pattern: wire Jev as a hook or extension, not as a skill the main model has to choose to call.

LangChain

TypeSafe-related LangChain middleware is experimental and evolves independently from the core SDK. Pin the installed package version, follow that release's reference example, and test its tool and model configuration before using it for a gate or router. The model routing and confidence gating pages describe the framework-independent policy to preserve.

Vercel AI SDK

The Vercel path exposes Jev as an evaluation model rather than a chat model:

import { typeSafeAi } from '@ai-sdk/typesafe-ai'
import { experimental_evaluate } from 'ai'

const result = await experimental_evaluate({
  model: typeSafeAi.evaluationModel('jev-latest'),
  // ...
})

Two details worth knowing:

• In this path, a yes/no judgment is a boolean with a probability field, not a noul object. TypeSafe's confidence for a Choice or Score is at result.providerMetadata.typesafe.confidence[questionId].

For sensitive state, confirm the chosen provider's current data-handling terms before sending production data.

Gateways and Hosts

Gateway catalogs and model IDs change frequently. If you use a gateway, verify Jev's current availability, request format, data handling, and pricing in that gateway's documentation rather than assuming parity with direct access.

MCP Servers

Community MCP servers expose Jev as tools for MCP hosts:

• itsmostafa/typesafe-mcp is a community Go server. Review its source, versioning, and operational model before giving it credentials or production data.

See the MCP section for the server format and host setup.

Integration Checklist

1. Pin the model. Request jev-1.13.0 and log the versioned ID from every response.
2. Keep the key server-side. Never ship it to a browser or log it.
3. Own the thresholds. Prompts and criteria live in one reviewable file; thresholds and actions live in code.
4. Shadow first. Log decisions before they can block or execute.
5. Log the decision surface. Chosen option, confidence, model, and usage — so the policy is auditable.
6. Route as a hook, not a skill. The main model should not pay to invoke the router.

Related

• Getting Started — SDK install and first call
• Model Routing — a policy these integrations can support
• Tool Comparison — how the coding harnesses in this section handle routing and gating