Agent tooling

Jevbridge

@tacticocc★ 25TypeScriptMITupdated 2026-09-18▶ runnable

ACP and MCP adapter that bridges TypeSafe Jev with any LLM — computer use and typed decisions alongside Codex, Claude, Grok, and OpenCode.

tacticocc/Jevbridge ↗

Where it calls Jev

export const TYPESAFE_ENDPOINT = "https://api.typesafe.ai/v1/systemone";

src/types.ts:111

The link points at the commit we read, so the line number still holds.

What it asks Jev

These question sets are lifted from this project's source exactly as written.

from src/mcp/tools.test.ts:30

  1. refundtrue/false

    Does this request a refund?

  2. teamchoice

    Which team?

    • billing — Payments, invoices, refunds.
    • technical — Bugs and outages.
    • other — None of the above.

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    refund: { type: 'boolean', instructions: 'Does this request a refund?' },
    team: { type: 'choice', instructions: 'Which team?',
      criteria: { billing: 'Payments, invoices, refunds.', technical: 'Bugs and outages.', other: 'None of the above.' } },
  },
});

Other projects in this category