typesafe-jev

TypeSafe Jev — integration reference

Verified against docs.typesafe.ai, the OpenAPI spec, PyPI, npm and the integration docs on 2026-10-03. Versions at that date: typesafe-sdk 0.7.2, @typesafe-ai/sdk 0.6.0, model jev-1.13.0, Pydantic AI 2.53.0, @ai-sdk/typesafe-ai 3.0.12, langchain-typesafe 0.0.1a3 (alpha). The SDKs are pre-1.0 and have shipped breaking changes in minor releases — pin them.

Deeper material:

Deep Knowledge: Use mcp__documentation__fetch_docs with technology: typesafe-jev and one of these topics:

• api-reference: every field, error body and limit, plus where the docs and the OpenAPI spec disagree.
• python-sdk and javascript-sdk: the full SDK surface, mock-transport tests and concurrency patterns.
• pydantic-ai, langchain, vercel-ai-sdk, dspy: one page per integration.
• recipes-extraction and recipes-routing: the official cookbooks, ported and runnable.
• evidence: every independent study, with its methodology and caveats.

Every code block there was quoted from a cited source, executed offline against mocks, or type-checked.

Mental model

One request = one state (what is judged) + a map of named questions (what is asked about it). Every question is answered independently against the same state, in parallel. The question keys are yours and are not sent to the model — only instructions and criteria are.

POST https://api.typesafe.ai/v1/systemone
Authorization: Bearer $TYPESAFE_API_KEY

{
  "state": "Help! My payouts have been failing for 3 days.",
  "model": "jev-latest",
  "questions": {
    "is_urgent": {"type": "noul", "instructions": "Does this convey urgency?",
                  "criteria": {"true": "Explicitly time-sensitive", "false": "No urgency expressed"}},
    "department": {"type": "choice", "instructions": "Which team should handle this?",
                   "criteria": {"billing": "Payments, invoicing, refunds",
                                "technical": "Bugs, outages, integrations",
                                "sales": "Pricing, upgrades, new accounts"}},
    "frustration": {"type": "score", "instructions": "How frustrated is the customer?",
                    "criteria": ["Calm", "Frustrated", "Very angry"]}
  }
}