moeru-ai/airi · error · Error

Failed to download MediaPipe vision task asset for

Error message

Failed to download MediaPipe vision task asset for ${key} after ${attempt} attempts

What it means

prepare-tasks.ts is the package's asset-preparation script (run as a task before dev/build) that downloads MediaPipe vision task model files (pose/hands/face .task assets) into the package's assets directory. Each downloadAsset call wraps a retrying downloader; when every attempt fails (network error, DNS, proxy block, 4xx/5xx), it rethrows with this message and the last underlying error attached as cause. The script skips already-present usable files, so this only fires when the asset genuinely cannot be fetched.

Solutions

Fix network egress for the model CDN the script downloads from, then re-run the task — the retry wrapper will try again from scratch.
Check the console.warn lines emitted per attempt (they include the underlying errorMessageFromValue) to identify DNS vs HTTP status vs proxy errors.
Pre-seed the asset: place a valid, non-empty .task file at the expected outputPath (printed in the error context / taskTargets) so the isUsableFile check skips the download.
If you maintain the package, verify each taskTargets source URL still resolves (curl -I) and update it.

Defensive patterns

Strategy: retry

Prevention

Cache the prepared assets directory on CI so model downloads rarely re-run.
Pre-seed assets in air-gapped environments by copying a known-good .task file to the expected outputPath (script skips existing usable files).
Watch the per-attempt console.warn lines to distinguish DNS, proxy, and HTTP-status failures before re-running.

When it happens

Trigger: Running pnpm -F @proj-airi/model-driver-mediapipe dev/build (or any pipeline invoking tasks/prepare-tasks.ts) on a machine without internet access; behind a corporate proxy that blocks jsdelivr/Google-hosted model files; a mistyped or moved upstream model URL in taskTargets; CI runner with flaky egress.

Common situations: Offline development or air-gapped CI; corporate MITM proxies rejecting large binary downloads; the upstream CDN URL changed between package versions; intermittent network during long CI runs.

Related errors

AI-assisted analysis of moeru-ai/airi@677329427f (2026-08-18). Data as JSON: /api/errors/50d4b01adc73fada. Report an issue: GitHub.

Appendix: source

Thrown at packages/model-driver-mediapipe/tasks/prepare-tasks.ts:75

      await fs.writeFile(tempPath, Buffer.from(response))
      await fs.rename(tempPath, outputPath)
      console.log(`MediaPipe vision task asset for ${key} saved to ${outputPath}`)
    }
    finally {
      await fs.rm(tempPath, { force: true })
    }
  }, {
    onError: (error) => {
      const message = errorMessageFromValue(error)
      console.warn(`Failed to download MediaPipe vision task asset for ${key} (attempt ${attempt}): ${message}`)
    },
  })

  try {
    await downloadWithRetry()
  }
  catch (error) {
    throw new Error(`Failed to download MediaPipe vision task asset for ${key} after ${attempt} attempts`, {
      cause: error,
    })
  }
}

await fs.mkdir(assetsRoot, { recursive: true })

for (const { key, source, outputPath } of taskTargets) {
  if (await isUsableFile(outputPath)) {
    console.log(`MediaPipe vision task asset for ${key} already exists at ${outputPath}, skipping download.`)
    continue
  }
  await downloadAsset(key, source, outputPath)

  if (!await isUsableFile(outputPath))
    throw new Error(`Failed to ensure MediaPipe vision task asset for ${key}: missing or empty file at ${outputPath}`)
}

View on GitHub (pinned to 677329427f)