Agent tooling

pijev

@tonyzdev★ 2TypeScriptMITupdated 2026-09-20▶ runnable

PiJev: a terminal coding agent with Jev in the loop — Jev ranks the repository's files before the first call, picks skills and triages failures; your coding model writes the code. Built on Pi.

tonyzdev/pijev ↗

Where it calls Jev

const endpoint = env.PIJEV_JEV_ENDPOINT ?? (provider === "vercel" ? "https://ai-gateway.vercel.sh/v4/ai" : "https://api.typesafe.ai/v1/systemone");

src/config.ts:26

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 test/gateway.test.ts:7

  1. relevanttrue/false

    Is this relevant?

  2. categorychoice

    Choose a category.

    • code — Code defect
    • environment — Missing executable

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: {
    relevant: { type: 'boolean', instructions: 'Is this relevant?' },
    category: { type: 'choice', instructions: 'Choose a category.',
      criteria: { code: 'Code defect', environment: 'Missing executable' } },
  },
});

from test/jev.test.ts:7

  1. categorychoice

    Classify the failure.

    • code — Code defect
    • network — Network failure

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: {
    category: { type: 'choice', instructions: 'Classify the failure.',
      criteria: { code: 'Code defect', network: 'Network failure' } },
  },
});

Other projects in this category