@mcp-abap-adt/typesafe-decision

TypeSafe AI (Jev) decision-model provider for @mcp-abap-adt/llm-agent.

Jev is not an LLM: it answers typed questions (noul yes/no, choice, score) about a state with numbers only. This package implements IDecisionModel over @typesafe-ai/sdk.

Exports:

• TypeSafeDecisionModel — implements IDecisionModel.
• TypeSafeDecisionConfig — configuration type.

import { staticApiKey } from '@mcp-abap-adt/llm-agent';
import { TypeSafeDecisionModel } from '@mcp-abap-adt/typesafe-decision';

const model = new TypeSafeDecisionModel({
  credential: staticApiKey(process.env.DECISION_API_KEY ?? ''),
});
const r = await model.decide({
  state: 'I was charged twice.',
  questions: { billing: { type: 'noul', instructions: 'Is this about billing?' } },
});
if (r.ok) console.log(r.value.answers.billing);

The key is asked from the credential on every call, so a rotating key rotates. TYPESAFE_* environment variables are never read: every client option is passed explicitly. Data sent: the state and the questions go to TypeSafe's API.

License

GNU Lesser General Public License v3.0 only (LGPL-3.0-only) — see LICENSE (LGPL) and GPL-3.0.txt (the GPL it layers permissions onto; both are required, the LGPL is not standalone).

Copyright © 2025–2026 Oleksii Kyslytsia

Importing this package, or running it behind an HTTP endpoint, does not place your program under the LGPL — the licence asks that modifications to this library stay free and that your users can substitute their own build. Versions up to and including v20.9.5 were MIT and stay MIT; the change is not retroactive. Full detail, including what to do if you redistribute: docs/LICENSING.md.