Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.
Python agentagent-osaiai-agentsai-partnerchatopsclaudeclaude-codecodexdevtoolsdiscord-botlark-botllmlocal-firstlong-term-memoryopencodeslack-botvibe-codingvibeywechat
import { readFileSync } from 'node:fs';
import { dirname, resolve } from 'node:path';
import { fileURLToPath } from 'node:url';
import { describe, expect, it } from 'vitest';
import { EFFORT_BY_BACKEND, REASONING_EFFORTS } from '@/lib/effortOptions';
import { TIER_SUGGESTIONS } from './tierSuggestions';
import { MODEL_PROTOCOLS } from './types';
const HERE = dirname(fileURLToPath(import.meta.url));
const BACKEND_MODELS = resolve(HERE, '../../../../..', 'vibe/data/backend_models.json');
const E2E_PROTOCOL_TIERS = resolve(HERE, '../../../../e2e/support/protocol-tiers.json');
/**
* Tiers the builtin catalog names that the unified vocabulary does not.
*
* Empty on purpose: the vocabulary is the ordered SUPERSET of every value a
* catalog row legitimately declares (`ultra` included, because gpt-5.6-sol/terra
* name it). A NEW catalog-only tier fails this equality rather than reaching
* the UI unnamed; a future exemption is an explicit pin here, never a loosened
* assertion.
*/
const KNOWN_CATALOG_DIVERGENCE: readonly string[] = [];
const vocabulary = REASONING_EFFORTS as readonly string[];
const rank = (effort: string) => vocabulary.indexOf(effort);
const catalogTiers = (): Map<string, string[]> => {
const catalog = JSON.parse(readFileSync(BACKEND_MODELS, 'utf8'));
return new Map(
Object.entries(catalog.backends as Record<string, { models?: { reasoning_efforts?: string[] }[] }>)
.map(([backend, entry]) => [
backend,
(entry.models ?? []).flatMap((model) => model.reasoning_efforts ?? []),
]),
);
};
describe('unified reasoning-effort vocabulary', () => {
it('is the spec\'s ordered superset, including catalog-only ultra', () => {
// The UI-side pin of the frozen vocabulary. The cross-file mirror in
// `reasoningEffortVocabulary.mirror.test.ts` holds this list equal to the
// backend export when that name exists on the branch, and to this same
// frozen vocabulary while it does not.
expect([...REASONING_EFFORTS]).toEqual([
'none', 'minimal', 'low', 'medium', 'high', 'xhigh', 'max', 'ultra',
]);
});
it('offers one suggestion list per source protocol, drawn from the vocabulary in its order', () => {
expect(new Set(Object.keys(TIER_SUGGESTIONS))).toEqual(new Set(MODEL_PROTOCOLS));
for (const [protocol, tiers] of Object.entries(TIER_SUGGESTIONS)) {
if (protocol === 'google') expect(tiers).toEqual([]);
else expect(tiers.length, protocol).toBeGreaterThan(0);
expect(new Set(tiers).size, protocol).toBe(tiers.length);
expect(tiers.filter((tier) => vocabulary.includes(tier)), protocol).toEqual([...tiers]);
expect([...tiers].sort((a, b) => rank(a) - rank(b)), protocol).toEqual([...tiers]);
// Family defaults are rung-1 over-claim protection, not the catalog
// superset: an unknown relay must not be handed `ultra`.
expect(tiers, protocol).not.toContain('ultra');
expect(tiers, protocol).not.toContain('none');
}
});
it('keeps the agent-backend effort lists inside the same vocabulary and order', () => {
for (const [backend, efforts] of Object.entries(EFFORT_BY_BACKEND)) {
expect(efforts.filter((effort) => vocabulary.includes(effort)), backend).toEqual(efforts);
expect([...efforts].sort((a, b) => rank(a) - rank(b)), backend).toEqual(efforts);
}
});
it('says everything the builtin model catalog says, apart from the pinned divergence', () => {
const declared = new Set([...catalogTiers().values()].flat());
const divergent = [...declared].filter((tier) => !vocabulary.includes(tier));
expect(new Set(divergent)).toEqual(new Set(KNOWN_CATALOG_DIVERGENCE));
});
it('suggests for the Anthropic protocol exactly what the claude catalog declares', () => {
// The claim in tierSuggestions.ts is about THIS family only. The codex rows
// answer a different question — which levels the Codex CLI accepts, not
// which the OpenAI wire protocols name — so they are checked against the
// vocabulary above and not against `TIER_SUGGESTIONS.openai_*`.
const claude = catalogTiers().get('claude') ?? [];
expect(claude.length).toBeGreaterThan(0);
expect(new Set(claude)).toEqual(new Set(TIER_SUGGESTIONS.anthropic));
});
it('is the same table the Playwright suite asserts against the live editor', () => {
// The e2e tsconfig cannot import `@/` aliases, so the suite reads this JSON
// rather than `TIER_SUGGESTIONS` itself. Equality here is what keeps that
// copy from silently describing a vocabulary the product no longer offers.
const e2e = JSON.parse(readFileSync(E2E_PROTOCOL_TIERS, 'utf8')) as Record<string, string[]>;
expect(e2e).toEqual(TIER_SUGGESTIONS);
});
});