Themis

A unit test framework built for AI coding agents.

Drop-in alternative to Jest and Vitest. Agents write tests, get structured failure output, and self-repair — all in the same edit-test-fix loop.

• Faster — 68.59% faster than Vitest, 130.26% faster than Jest on the same benchmark (proof)
• Agent-native — --agent JSON with failure clusters and structured repair hints
• One-command migration — npx themis migrate jest or vitest with codemods
• Modern by default — .ts, .tsx, .js, .jsx, ESM, React Testing Library, no config gymnastics
• Discoverable — ships AGENTS.md, themis.ai.json, and a Tessl tile so agents find and adopt it automatically

---

Quickstart

npm install -D @vitronai/themis@latest
npx themis init --agents
npx themis generate src     # or `app` for Next App Router repos
npx themis test

init --agents writes config, updates .gitignore, and scaffolds a downstream AGENTS.md.

Using Claude Code? Run npx themis init --claude-code to install CLAUDE.md, a Claude Code skill, and slash commands (/themis-test, /themis-generate, /themis-migrate, /themis-fix).

---

Performance

On the same React showcase benchmark, Themis measured 68.59% faster than Vitest and 130.26% faster than Jest on median wall-clock time. The comparison artifact is emitted by CI as .themis/benchmarks/showcase-comparison/perf-summary.json.

The first-try benchmark measures how often an AI agent generates tests that pass on the first run — the metric that matters most for agent-driven development:

ANTHROPIC_API_KEY=sk-... npm run benchmark:first-try

---

How it works

Plain English → structured tests

intent('user can sign in', ({ context, run, verify, cleanup }) => {
  context('a valid user', (ctx) => {
    ctx.user = { email: '[email protected]', password: 'pw' };
  });

  run('the user submits credentials', (ctx) => {
    ctx.result = { ok: true };
  });

  verify('authentication succeeds', (ctx) => {
    expect(ctx.result.ok).toBe(true);
  });

  cleanup('remove test state', (ctx) => {
    delete ctx.user;
  });
});

Phase names: context, run, verify, cleanup. Legacy aliases (arrange/act/assert, given/when/then) also supported.

Mocks and UI primitives

mock('../src/api', () => ({
  fetchUser: fn(() => ({ id: 'u_1', name: 'Ada' }))
}));

const { fetchUser } = require('../src/api');

test('mock captures calls', () => {
  const user = fetchUser();
  expect(fetchUser).toHaveBeenCalledTimes(1);
  expect(user).toMatchObject({ id: 'u_1', name: 'Ada' });
});

For jsdom tests, Themis ships render, screen, fireEvent, waitFor, useFakeTimers, mockFetch, and more. Full list in the API reference.

Code generation

Themis scans your source tree and generates contract tests for exported modules, React components, hooks, Next.js routes, and services:

npx themis generate src
npx themis test

When generated tests fail:

npx themis test --fix

--fix regenerates affected targets and reruns the suite. See the API reference for all generation flags (--review, --plan, --write-hints, --strict, --changed, etc.).

Migration

npx themis migrate jest
npx themis migrate vitest
npx themis test

One command scaffolds a compatibility bridge. Add --rewrite-imports to rewrite import paths, --convert for codemods. See the migration guide.

---

Config

themis.config.json:

{
  "testDir": "tests",
  "generatedTestsDir": "__themis__/tests",
  "testRegex": "\\.(test|spec)\\.(js|jsx|ts|tsx)$",
  "maxWorkers": 7,
  "reporter": "next",
  "environment": "node",
  "setupFiles": ["tests/setup.ts"],
  "tsconfigPath": "tsconfig.json"
}

Use environment: "jsdom" for DOM-driven tests. Themis auto-stubs common style/asset imports (.css, .scss, .png, .svg, etc.).

---

TypeScript

{
  "compilerOptions": {
    "types": ["@vitronai/themis/globals"]
  }
}

Ships first-party typings for all test APIs, typed intent context, and project-aware module loading for .ts, .tsx, ESM .js, .jsx, and tsconfig path aliases.

---

Pair with Alethia

Themis owns the unit/contract layer. Alethia owns the E2E/policy layer. Together they form the tightest test loop an autonomous coding agent can sit inside:

1. Agent generates code
2. Themis verifies the contract in milliseconds
3. Alethia verifies the running app in a real browser, under safety policy, with a signed audit trail

Use Themis on its own — it's MIT and stands alone.

---

Reference docs

• API reference — all CLI flags, globals, matchers, mocks, UI primitives
• Agent adoption guide — downstream repo setup
• Migration guide — Jest/Vitest migration details
• Why Themis — positioning and differentiators
• Showcase comparisons — direct Jest/Vitest examples
• Tutorial: Testing with Claude Code
• VS Code extension
• Release policy
• Publish guide
• Changelog