Batch Processing

💡TL;DR

Notemd processes entire folders in one action with configurable concurrency and overwrite control. Right-click a folder to batch-add wiki-links, extract concepts, research, or translate all notes within. Concurrency limits prevent API rate-limit errors. Progress is reported per file. Overwrite behavior is configurable: skip existing, append, or replace. Failed files are logged without aborting the batch.

This is part of the Obsidian AI Knowledge Management Guide.

Overview

Batch processing turns a folder of notes into a single operation. Instead of opening each note and running commands individually, you right-click the folder and select the task. Notemd iterates through every .md file, applies the chosen action, and reports progress in real time.

This feature is essential for vault-wide knowledge extraction. After importing dozens of PDFs, for example, batch-add-links followed by batch-extract-concepts builds your knowledge graph in minutes rather than hours.

How It Works

Batch Execution Model

1. File collection -- Notemd scans the target folder recursively (or top-level only, depending on settings) and collects all .md files.
2. Concurrency partitioning -- Files are divided into chunks based on the batchConcurrency setting. Each chunk runs in parallel; chunks run sequentially.
3. Execution -- Each file is processed using the same logic as the single-file command. Per-task provider and model settings are respected.
4. Progress reporting -- A toast notification updates after each file completes, showing N / Total progress.
5. Error handling -- If a file fails (API error, network timeout, etc.), the error is logged and the batch continues. The final summary lists any failed files.
6. Completion -- A summary toast reports total processed, successes, and failures.

Overwrite Behavior

When processing a file that already has wiki-links, concept notes, or translations, Notemd's behavior depends on the overwrite setting:

Mode	Behavior
Skip	Existing content is left untouched. Only unmodified files are processed.
Append (default)	New content is appended. Existing wiki-links, concepts, or translations are preserved.
Replace	The file is fully reprocessed. All previous Notemd modifications are overwritten.

For wiki-linking specifically: if a note already contains [[wiki-links]], the skip mode leaves it alone, while replace re-sends the entire note to the LLM for fresh link insertion. Use skip for incremental processing and replace for reprocessing after a model upgrade.

Concurrency Control

The batchConcurrency setting limits parallel API calls. This prevents rate-limit errors (HTTP 429) when processing large folders against providers with strict quotas.

Concurrency	Recommended For	Typical Rate-Limit Impact
1	Free tiers, strict providers	None (serial)
3 (default)	Most cloud providers	Low
5	Ollama (local), generous tiers	None / Low
10	Local models with fast inference	None

If you encounter 429 errors during batch processing, reduce concurrency to 1 or 2.

Configuration

Setting	Default	Effect
batchConcurrency	3	Maximum parallel API calls during folder operations
batchOverwriteExisting	false	Overwrite existing Notemd content. false = append mode.
batchSkipProcessed	false	Skip files that already contain Notemd markers (e.g., wiki-links)
batchRecursive	true	Include subdirectories when scanning the folder
enableStableApiCall	false	Enable retry logic (up to 4 attempts) per file during batch

Per-Task Models in Batch

Each batch operation uses the corresponding per-task model. Batch-add-links uses addLinksProvider, batch-research uses researchProvider, and so on. This means you can assign cheap models for high-volume operations and reserve expensive models for quality-sensitive tasks.

Example

You have a folder papers/ containing 40 imported research notes. You want to add wiki-links and extract concepts across all of them:

1. Right-click the papers/ folder
2. Select "Notemd: Process folder (add links)"
3. Notemd scans the folder, finds 40 .md files, and processes 3 at a time (default concurrency)
4. A progress toast shows: 12/40 files processed...
5. After ~3 minutes, a summary toast reports: 39 succeeded, 1 failed (API timeout on paper-37.md)
6. Repeat with "Notemd: Process folder (extract concepts)" to create concept notes for all 40

The one failed file is logged. You can re-run on just that file afterward.

Tips

• Start with low concurrency -- If you are unsure of your provider's rate limits, begin with 1 and increase gradually.
• Use skip mode for incremental updates -- After the first full batch, switch to batchSkipProcessed: true so only new notes get processed on subsequent runs.
• Enable stable API calls -- enableStableApiCall: true adds retry logic that recovers from transient network errors during long batches.
• Re-run after model upgrades -- If you switch to a better model, set batchOverwriteExisting: true and re-run to get improved links and concepts.

---

Next Steps

• Workflows -- Chain batch tasks into one-click sidebar buttons
• Custom Prompts -- Customize prompts for batch extraction
• Troubleshooting -- Fix rate-limit errors and connection failures during batch runs
• LLM Providers -- Per-task model configuration reference