Batch Processing Guide: Run AI Across Hundreds of Inputs at Once

# Batch Processing Guide: Run AI Across Hundreds of Inputs at Once

If you've ever found yourself copy-pasting the same prompt over and over — swapping out a product name here, a company name there — you already know the problem. It's slow, error-prone, and it completely defeats the point of having powerful AI at your fingertips.

Vincony's batch tools exist to solve that problem. Batch lets you fire off up to 50 independent prompts in a single run. CSV Batch goes one step further: upload a spreadsheet, write one template, and every row becomes its own AI request automatically. Both tools are available on Pro, Power, and Business plans, and both share the same credit system — you pay per prompt processed, just like a regular AI query.

This guide covers when to reach for each tool, how they actually work under the hood, and a practical walkthrough of both.

---

When to Use Batch Processing

Batch tools are the right choice whenever you have structured, repetitive work that follows the same pattern but varies by input. Common scenarios include:

• Content at scale — writing product descriptions, meta titles, or social captions for a whole catalogue
• Research and analysis — summarising 30 competitor pages, scoring 40 customer reviews by sentiment, or extracting key facts from a list of news headlines
• Personalised outreach — generating tailored email openers, icebreakers, or follow-up lines for a list of contacts
• Data enrichment — classifying rows in a CRM export, translating a list of strings, or tagging a library of items by category
• QA and review — running the same evaluation prompt against multiple drafts or outputs

The key signal: if you would naturally open a spreadsheet to track your inputs, CSV Batch is almost certainly the faster path. If you have a set of one-off prompts that don't share a common template, the plain Batch tool is simpler.

---

How It Works

Both tools send each prompt independently to the AI model you choose. There is no cross-contamination between items — each prompt is a fresh, isolated request. The engine processes them in parallel and streams results back as a single batch response, so the total wall-clock time is much shorter than running them sequentially yourself.

Credit cost is straightforward: one prompt costs the same number of credits as the model's standard per-query rate. You can see the estimated total in the UI before you hit Run Batch — it updates live as you add prompts or change the model.

Both tools require a Pro plan or higher. Free and Starter accounts do not have access to batch endpoints. You can check your current plan and remaining credits at pricing.

---

Tool 1: Batch Generation

Batch Generation is the quick-start option. No files to prepare — just type or paste your prompts directly into the browser.

Step-by-step

1. Open Batch Generation

Navigate to /os/batch. You'll see the Prompts card with a large text area.

2. Enter your prompts, separated by ---

Each block of text separated by three dashes on its own line becomes one independent prompt. The UI shows a live counter (X / 50 prompts) so you always know where you stand. If you go over 50, the Run Batch button blocks and shows an error.

For example, a tagline batch might look like:

Write a one-sentence tagline for a specialty coffee brand targeting remote workers --- Write a one-sentence tagline for a B2B SaaS startup in the logistics space --- Write a one-sentence tagline for a fitness app focused on busy parents

Not sure where to start? Click Try sample prompts in the empty state and the text area pre-fills with a working example you can modify.

3. Choose a model

Use the model selector (defaults to Gemini Flash) to pick the right balance of speed, quality, and credit cost. Each model shows its cost in credits next to its name. The estimated total cost updates immediately when you switch models.

4. Run the batch

Click Run Batch or press Ctrl+Enter (or Cmd+Enter on Mac) to start. A progress bar appears while the batch is running, along with an elapsed timer in seconds so you can gauge how long to expect. The browser tab title also changes to "Generating…" so you know it's working even if you've tabbed away.

5. Review results

Each result card shows the original prompt and the AI response. Long responses are collapsed to four lines by default — click Show more to expand. Hover any card to reveal a copy icon that copies just that response to your clipboard.

6. Export

Once results are in, you have two export paths:

• CSV — downloads a file with columns Prompt, Response, Credits, and Error (populated only for any failed items)
• Export & Share — opens a menu with direct integrations: Google Drive, Notion, Dropbox, Google Docs, OneDrive, Google Sheets, or email via Composio

Your draft auto-saves as you type, so if you close the tab and come back, your prompts are still there.

---

Tool 2: CSV Batch Generation

CSV Batch Generation is designed for structured data. Instead of writing each prompt manually, you define one template and the tool fills in the blanks from your spreadsheet.

Step-by-step

1. Prepare your CSV

Your file needs a header row. Column names become the variables you reference in your template. There's no required naming convention — use whatever headers make sense for your data. Maximum 50 data rows (not counting the header).

A simple product catalogue CSV might have columns: name, category, key_feature. A contact list might have: first_name, company, role.

2. Upload the file

Open /os/csv-batch and either click the dashed upload area or drag your .csv file directly onto it. The tool parses the file client-side, then shows:

• A confirmation badge with the row and column counts
• A clickable tag for every column header (e.g., {{name}}, {{category}})
• A preview table showing the first three rows of your data

If you spot a problem — wrong file, wrong encoding — click the X button to clear and re-upload.

3. Write your prompt template

In the Prompt Template card, write your prompt using {{column_name}} placeholders wherever you want the tool to inject a row's value. You can click any column tag from the upload card to insert the placeholder at the cursor position automatically.

A product description template might read:

Write a 2-sentence product description for {{name}}, a {{category}} product. Highlight that it {{key_feature}} and keep the tone friendly and direct.

As you type, a Preview (Row 1) panel beneath the text area renders what your template looks like with the first row's actual values substituted in. This is the fastest way to catch placeholder typos before you burn credits on a full run.

4. Choose a model and run

Select your model from the dropdown — it shows the credit cost per row next to each option. The row count and estimated total credit spend are displayed inline. Click Run Batch or Cmd/Ctrl+Enter to start. The same progress bar and elapsed timer from the plain Batch tool appear here too.

5. Review row-by-row results

Each result card is labelled Row N and shows the rendered prompt (with placeholders filled in) plus the AI response. Long responses collapse to four lines with a Show more toggle. Any rows that errored show a red alert icon instead of a response.

6. Export with original columns intact

The CSV export for this tool is richer than the plain Batch export. It writes back all your original columns plus three new ones appended on the right: AI_Response, Credits, and Error. That means you can open the output in Excel or Google Sheets and have a fully enriched version of your original data — no manual merging required.

The Export & Share menu offers the same cloud destinations as the plain Batch tool: Google Drive, Notion, Dropbox, Google Docs, Google Sheets, OneDrive, and email.

---

Tips for Getting the Most Out of Batch

Front-load context in your template. Each prompt is independent, so any background the model needs must be in the prompt itself. Don't assume it remembers anything from a previous row.

Use the Row 1 preview before running. It takes two seconds and catches the most common mistake — a mistyped placeholder like {{product_name}} when your column is actually named name.

Start with a cheaper model. Run your first pass on a fast, low-cost model (Gemini Flash is the default for good reason). Upgrade to a higher-quality model only for the rows that need a rerun.

Keep prompts focused. Shorter, specific prompts tend to outperform vague ones in batch contexts. If your template is three paragraphs of preamble, trim it.

Use the error column. If a run partially fails, the exported CSV flags exactly which rows errored. You can filter those rows, fix the underlying data, and re-run just that subset rather than repeating the whole batch.

Draft auto-save is your friend. Both tools persist your inputs (prompts or template) across page reloads. If you close the browser mid-edit, your work will be waiting when you come back.

---

Credits and Plans

Both Batch and CSV Batch are Pro plan and above features. On Power and Business plans you get larger monthly credit allocations, which makes batch workloads significantly more economical at scale. See pricing for current plan limits and per-model credit costs.

If a batch run would exceed your available credits, the tool blocks the run before sending any requests and shows an insufficient-credits error — you won't be partially charged.

---

What to Do With Your Results

Batch outputs pair naturally with other Vincony tools. If you're generating structured data like extracted fields or classifications, the Data Analyzer can take your exported CSV and help you spot patterns. If you're building a longer content workflow, the Auto-Pilot agent can handle multi-step tasks that go beyond a single prompt pass. For pulling structured data out of raw text before you batch it, the Extractor is worth a look.

The fastest way to start is to take a spreadsheet you already have — a product list, a contact export, a content calendar — and drop it into CSV Batch. Write one template sentence, run it against the first five rows, and see what comes back. Most teams find a working workflow within the first ten minutes.