Help Center

How can we help you?

Find answers to common questions about BatchForge, learn best practices, and get the most out of your batch processing workflows.

Getting Started
New to BatchForge? Start here to learn the basics.
How Batch Processing Works
Learn how BatchForge processes your data at scale.
Billing & Credits
Understand how credits work and manage your usage.
Contact Support
Get help from our team.

Getting Started

Create your first batch job

BatchForge makes it easy to process large amounts of data with AI models. Here's how to get started:

  1. Upload a CSV file with the data you want to process
  2. Write a prompt template using {{variables}} from your columns
  3. Select an AI model (GPT-4o, Claude, Gemini, etc.)
  4. Run your job and download results when complete

Understanding processing modes

BatchForge offers two processing modes:

  • Immediate ModeResults are streamed in real-time. Best for smaller jobs (up to 10,000 rows) when you need results quickly.
  • Batch Mode (50% off)Jobs are queued and processed asynchronously. Supports up to 500,000 rows and costs 50% less than immediate mode.

Using your own API keys (BYOK)

Pro and Business plan users can add their own API keys for any supported provider (OpenAI, Anthropic, Google, DeepSeek, Mistral, xAI). This allows unlimited processing without using platform credits. Your keys are encrypted and stored securely.

How Batch Processing Works

Upload and preview

Start by uploading your data file. BatchForge supports CSV files with headers. After upload, you'll see a preview of your data with detected columns.

  • Supported formats: CSV and JSON files
  • Column detection: Headers are automatically detected from the first row
  • Preview: See the first 5 rows to verify your data looks correct
  • Row limits: Free tier supports up to 100 rows, paid plans support up to 500,000 rows

Prompt templates with variables

Write your prompt using variables that reference your CSV columns. Variables use double curly braces and will be replaced with the actual value for each row.

Example prompt template:

Analyze the following product review and determine the sentiment:

Product: {{product_name}}
Review: {{review_text}}

Return the sentiment as positive, negative, or neutral.

  • Save templates: Reuse prompts across multiple jobs
  • Variable highlighting: Valid variables are highlighted in green
  • Best practice: Be specific and include examples in your prompt for better results

Processing and progress tracking

Once you start a job, BatchForge processes your rows and shows real-time progress.

  • Progress bar: See how many rows have been processed
  • Live results: Watch results appear as each row completes
  • Pause/Resume: Pause processing at any time and resume later
  • Automatic retries: Failed rows are automatically retried up to 3 times
  • Error handling: Rows that fail after retries are marked with error details

Downloading results

When your job completes, download the results as a CSV file.

  • CSV export: Results are exported in standard CSV format
  • Original columns: Choose to include your original input columns alongside AI-generated outputs
  • Structured outputs: If you defined output columns, each gets its own column in the CSV
  • Re-download: Access and re-download results from any completed job in your history

Supported Models & Pricing

OpenAI models

OpenAI offers a range of GPT and reasoning models with different capabilities and price points.

  • GPT-5 / GPT-5.1 / GPT-5.2Flagship models for complex reasoning, nuanced analysis, and high-quality outputs. Available on Pro and Enterprise plans.
  • GPT-5 MiniBalanced cost and quality for most tasks. Available on Starter tier and above.
  • GPT-5 NanoHigh-volume simple transforms at low cost. Available on Free tier and above.
  • GPT-4.1 / GPT-4.1 Mini / GPT-4.1 NanoCode-focused models optimized for programming and structured output tasks.
  • o4-mini / o3-miniReasoning models designed for math, logic, and code tasks that benefit from chain-of-thought.

Anthropic models (Claude)

Claude models excel at nuanced understanding, long documents, and following complex instructions.

  • Claude Opus 4.5Deep analysis and complex reasoning. Available on Enterprise plans.
  • Claude Sonnet 4.5Nuanced writing and analysis with excellent instruction-following. Available on Pro tier and above.
  • Claude Haiku 4.5Fast, high-quality responses at low cost. Available on Starter tier and above.
  • Claude Opus 4 / Sonnet 4 / Haiku 4Previous generation models, still available for all use cases.
  • Claude 3.5 Sonnet / HaikuLegacy models, still supported for existing workflows.

Google models (Gemini)

Google's Gemini models offer strong performance with competitive pricing.

  • Gemini 3 ProComplex tasks with web grounding capabilities. Available on Pro tier and above.
  • Gemini 2.5 ProAdvanced reasoning with built-in search capabilities. Available on Pro tier and above.
  • Gemini 2.5 Flash / Flash-LiteFast and affordable for everyday processing. Available on Starter tier and above.
  • Gemini 2.0 Flash / Flash-LiteCheapest option for high-volume batch processing. Available on Free tier and above.

DeepSeek, Mistral & xAI

BatchForge also supports models from DeepSeek, Mistral, and xAI for additional flexibility and pricing options.

DeepSeek

  • DeepSeek Chat (V3.2)Affordable general-purpose model. Available on Starter tier and above.
  • DeepSeek R1Complex reasoning and math tasks with chain-of-thought. Available on Starter tier and above.

Mistral

  • Mistral Large 3Complex multilingual tasks. Available on Starter tier and above.
  • Mistral Medium 3.1Balanced performance with strong European language support. Available on Starter tier and above.
  • Mistral Small 3.2Fast everyday processing at low cost. Available on Free tier and above.
  • Mistral NemoUltra-high-volume simple tasks at minimal cost. Available on Free tier and above.

xAI

  • Grok 4Advanced reasoning capabilities. Available on Pro tier and above.
  • Grok 3High-performance general-purpose model. Available on Pro tier and above.
  • Grok 3 MiniAffordable everyday processing. Available on Starter tier and above.

Pricing and cost estimation

Costs vary by model and are based on token usage (input + output tokens). BatchForge shows you an estimated cost before you run each job.

  • Batch mode discount: All models cost 50% less in batch mode
  • Cost preview: See estimated credits before running a job
  • Usage tracking: Monitor your credit usage in the dashboard
View Full Pricing

Structured Outputs

Defining output schemas

Instead of getting free-form text from the AI, you can define exactly what columns you want in your output. BatchForge will instruct the model to return structured data that maps directly to your columns.

Example output columns:

• brand (text)

• category (text)

• sentiment_score (number)

• is_urgent (boolean)

Supported column types

  • text — String values (names, descriptions, categories)
  • number — Numeric values (scores, counts, prices)
  • boolean — True/false values (flags, indicators)
  • date — Date values (parsed from text)
  • json — Complex nested objects

Example: Product categorization

Automatically categorize products based on their names and descriptions.

Prompt:

Categorize this product:
Name: {{product_name}}
Description: {{description}}

Output columns:

• category (text) — Main category like "Electronics", "Clothing", "Home"

• subcategory (text) — More specific category

• confidence (number) — Confidence score from 0 to 100

Sample output:

categorysubcategoryconfidence
ElectronicsHeadphones95

Example: Sentiment analysis

Analyze customer feedback to extract sentiment and key insights.

Prompt:

Analyze the sentiment of this customer review:
{{review_text}}

Output columns:

• sentiment (text) — "positive", "negative", or "neutral"

• score (number) — Sentiment score from -1 to 1

• keywords (text) — Key topics mentioned, comma-separated

• is_urgent (boolean) — Whether the review requires immediate attention

Sample output:

sentimentscorekeywordsis_urgent
negative-0.7shipping, delay, refundtrue

Billing & Credits

How credits work

Credits are deducted based on the actual token usage of your jobs. Each model has different costs per 1,000 tokens. You can see the estimated cost before running a job.

New users receive 50 free credits ($0.50 value) to try BatchForge.

Batch mode savings (50% off)

Batch mode jobs are processed asynchronously and cost 50% less than immediate mode. This is ideal for large jobs that don't require real-time results.

Subscription plans

Choose a plan that fits your processing needs.

  • Free50 credits/month. Access to free-tier models. Perfect for trying BatchForge.
  • Starter ($9.99/month)1,000 credits/month. Access to all LLM providers and models.
  • Pro ($49.99/month)5,000 credits/month. BYOK support and priority processing included.
  • Business ($299.99/month)30,000 credits/month. Team management, user roles & permissions, BYOK, and dedicated support.
  • EnterpriseCustom credit allocation, API access, SLA guarantee, and dedicated support. Contact us for pricing.

Annual plans are available for Starter, Pro, and Business tiers — save 2 months free when you pay annually.

Using your own API keys (BYOK billing)

With BYOK (Bring Your Own Keys), you pay the AI providers directly instead of using BatchForge credits.

  • No credit usage: Jobs using your own API keys don't consume BatchForge credits
  • Direct billing: You're billed directly by the provider (OpenAI, Anthropic, Google, DeepSeek, Mistral, or xAI)
  • Plan requirement: BYOK is available on Pro and Business plans
  • Secure storage: Your API keys are encrypted and stored securely
  • Multiple keys: Add keys for all six providers (OpenAI, Anthropic, Google, DeepSeek, Mistral, xAI) and switch between them per job

Contact Support

Email Support

Have a question or need help? Reply to any BatchForge email or contact us at:

[email protected]

Feature Requests & Feedback

We'd love to hear your ideas for improving BatchForge.

Share Feedback

Bug Reports

Found something that's not working right? Let us know so we can fix it.

  • • Describe what happened and what you expected
  • • Include the job ID if related to a specific job
  • • Screenshots or error messages are helpful
Report a Bug