janhq/jan · error · Error

Backend setup was not successful. Please restart the app in

Error message

Backend setup was not successful. Please restart the app in a stable internet connection.

What it means

Thrown by getDevices() after splitting this.config.version_backend on '/'. The setting must be in the form '<version>/<backend>' (e.g. 'b4000/vulkan'). If either half is empty/falsy, the backend was never resolved, so device enumeration cannot proceed. The misleading 'stable internet connection' text points at the root cause: the backend download/selection step that populates version_backend failed or did not complete.

Source

Thrown at extensions/llamacpp-extension/src/index.ts:3999

    if (!touched) return

    await invoke<void>('write_yaml', { data: cfg, savePath: configPath })

    try {
      await this.refreshRouterPreset()
    } catch (e) {
      logger.warn(
        `Failed to restart router after model settings update for ${modelId}`,
        e
      )
    }
  }

  async getDevices(): Promise<DeviceList[]> {
    const cfg = this.config
    const [version, backend] = cfg.version_backend.split('/')
    if (!version || !backend) {
      throw new Error(
        'Backend setup was not successful. Please restart the app in a stable internet connection.'
      )
    }
    // set envs
    const envs: Record<string, string> = {}
    if (this.llamacpp_env) this.parseEnvFromString(envs, this.llamacpp_env)

    // Ensure backend is downloaded and ready before proceeding
    await this.ensureBackendReady(backend, version)
    logger.info('Calling Tauri command getDevices with arg --list-devices')
    const backendPath = await getBackendExePath(backend, version)

    try {
      const dList = await invoke<DeviceList[]>('plugin:llamacpp|get_devices', {
        backendPath,
        envs,
      })
      // On Linux with AMD GPUs, llama.cpp via Vulkan may report UMA (shared) memory as device-local.

View on GitHub (pinned to fad3f12a14)

Solutions

1. Ensure the app had network access on first launch so the backend manifest downloaded and version_backend was populated.
2. Restart the app so the setup/init flow re-runs and rewrites version_backend.
3. Open settings and confirm the llamacpp backend (and version) is selected; if blank, re-select it.
4. Check logs for the backend download/ensure step failing and resolve the underlying fetch error (proxy, disk space, permissions).

Example fix

// before
const devices = await engine.getDevices()

// after
const [version, backend] = (engine.config.version_backend || '').split('/')
if (!version || !backend) {
  throw new Error('llamacpp backend not configured yet; finish first-run setup before listing devices')
}
const devices = await engine.getDevices()

function isVersionBackendConfigured(vb: string | undefined): boolean {
  if (!vb) return false
  const [v, b] = vb.split('/')
  return Boolean(v && b)
}

if (!isVersionBackendConfigured(engine.config.version_backend)) {
  throw new Error('Finish first-run backend setup before listing devices')
}

function isVersionBackend(v: unknown): v is `${string}/${string}` {
  return typeof v === 'string' && /^[^/]+\/+[^/]+$/.test(v)
}

try {
  return await engine.getDevices()
} catch (e) {
  if (/Backend setup was not successful/.test(String(e))) {
    await waitForBackendReady() // poll setup/init, then retry once
    return await engine.getDevices()
  }
  throw e
}

When it happens

Trigger: Calling getDevices() during or before first-run backend setup; version_backend setting is empty, undefined, or malformed (no '/' separator, or trailing/leading slash). Also when the settings migration that writes version_backend has not run for this install.

Common situations: Fresh install where the backend manifest fetch failed (offline/corporate proxy); a settings reset that wiped version_backend; a downgrade/upgraded build whose default version_backend key changed; corrupted persisted settings.

Related errors

• Checksum mismatch for ${name}; the download was corrupt or t
• Import of "${FALLBACK_EMBEDDING_MODEL_ID}" did not complete
• Failed to fetch supported backends: ${error instanceof Error
• ubatch_size (${ubatchSize}) is too small. Minimum required:
• Unable to find a suitable port for MLX model

AI-assisted analysis of janhq/jan@fad3f12a14 (2026-08-12). Data as JSON: /api/errors/e124be419b7e0473. Report an issue: GitHub.