AI Assistant

The AI Assistant is a conversational copilot built into Batcher.ai. It helps create data processing workflows through natural dialogue — no formula knowledge required. Describe what you need, and the assistant handles the rest: analyzing your data, proposing a plan, generating formulas, and injecting them directly into your spreadsheet.

Getting started

Opening the AI Assistant

Navigate to the AI Assistant from the main navigation. Each conversation creates a linked spreadsheet automatically. You can also link an existing spreadsheet to a new conversation.

Your first conversation

Simply describe what you want to accomplish in natural language. The assistant detects your language automatically and responds accordingly.

Examples of first messages:

“Translate column A into French”
“Generate SEO descriptions for each product”
“Compare ChatGPT and Perplexity responses for my data”
“I have a list of URLs, scrape them and extract the main content”

If your instruction is clear, the assistant executes immediately. For more complex tasks, it asks clarifying questions first.

How it works

Typical workflow

The assistant adapts to your needs, but a typical conversation follows this flow:

1. Understand your data

Import data into the spreadsheet (CSV, Excel, or manual entry), or ask the assistant to generate sample data if you want to explore first.

2. Describe your goal

Explain what you want in plain language. The assistant reads your spreadsheet context (headers, sample rows, column structure) and asks questions if needed.

3. Review the proposed plan

For multi-step workflows, the assistant proposes a structured plan with named steps. Each step shows:

Input columns used
Output column created
AI model suggested
Prompt template

You can modify the plan or approve it to proceed.

4. Formula injection

Once the plan is validated, the assistant generates formulas and injects them directly into the spreadsheet. Two execution modes are available:

Test on 3 rows — Apply formulas to the first 3 rows to verify results before committing
Full document — Apply formulas to all rows immediately

5. Iterate and refine

After formulas are applied, you can continue the conversation to adjust results, add more columns, change models, or start a new task on the same data.

Features

Real-time streaming

Responses appear word by word as the assistant generates them. Tool calls (reading the spreadsheet, writing formulas) are visible in real time so you always know what is happening.

Automatic language detection

The assistant detects the language of your message and responds in the same language. Your spreadsheet content can be in a different language — the assistant handles this naturally (e.g., it writes explanations in French while referencing English column headers as-is).

Expertise adaptation

The assistant adapts its communication style based on your vocabulary:

Beginner — Simple language, detailed explanations, concrete examples
Intermediate — Technical terms accepted, focus on efficiency
Expert — Concise responses, optimization focus, edge cases discussed

Suggested next actions

After each response, the assistant may suggest the next logical action as ghost text in the input field. Press Tab to accept the suggestion, or type your own message.

Undo support

Every write operation creates an automatic restore point. Ask the assistant to “undo” at any time to revert the last change. Multiple levels of undo are supported.

Model selection

Switch between AI models at any time using the model selector at the top of the chat. This changes which model the assistant uses for its responses — it does not affect the models referenced in your spreadsheet formulas.

Available models include the built-in default model and any AI configurations you have set up (ChatGPT, Claude, Gemini, etc.). See AI Providers & Configuration.

Conversation management

History — All conversations are saved automatically and accessible from the sidebar
Reset — Clear all messages while keeping the linked spreadsheet data
Redo — Delete the last assistant response and regenerate it
Delete — Permanently remove a conversation

What the assistant can do

Generate sample data

If you do not have data yet, ask the assistant to generate example data for your use case. The data is proposed first and loaded into the spreadsheet on confirmation.

Read and analyze your spreadsheet

The assistant sees your spreadsheet structure (headers, row count, sample data) and can read specific cells or ranges when needed to understand your data.

Write to the spreadsheet

The assistant can write headers, static content, and data directly into cells. All writes are reversible with undo.

Generate and inject AI formulas

The assistant creates =LLM() formulas with the appropriate prompts and model references, then injects them into the spreadsheet. Formulas are defined for row 2 and automatically expanded to all rows in the target range.

Propose workflow plans

For complex tasks, the assistant creates a step-by-step plan showing each processing stage, the columns involved, and the models to use. You review and approve before execution.

Manage spreadsheet structure

The assistant can insert or delete columns, move data, find and replace text, and reorganize your spreadsheet layout as part of building a workflow.

Working with formulas

When the assistant injects formulas, it follows these rules:

Formulas are always defined for row 2 (the template row)
The system automatically expands them to all rows in the target range (e.g., C2:C100)
Cell references adjust automatically: A2 in the template becomes A3, A4, A5, etc.
Absolute references ($A$1) remain fixed across all rows
Formula syntax matches the standard Batcher.ai format: =LLM(prompt, "config_name")

Example: The assistant might inject:

C2: =LLM(CONCAT("Translate to French: ", A2), "default")
Target range: C2:C150

This creates the formula in C2 through C150, with each row referencing its own data in column A.

Integrations available through the assistant

The assistant can create formulas using any AI provider you have configured:

Provider	What it does	Configuration
default	Built-in AI model, included free	No setup required
ChatGPT	OpenAI models (GPT-4o, etc.)	Requires API key
Claude	Anthropic models	Requires API key
Gemini	Google AI models	Requires API key
Perplexity	Search-augmented AI responses	Requires API key
Jina	Web page scraping — converts URL to markdown	Requires API key
Haloscan	SEO keyword data — returns search volumes, SERP data	Requires API key

The assistant automatically references your configured models in formulas. If no custom configurations exist, it uses "default" and mentions that you can add your own API keys for better results.

Tips for getting the best results

Be specific — “Generate a 50-word French product description for each item” works better than “describe my products”
Mention your columns — “Use column A as the product name and column B as the category” helps the assistant understand your data
Start simple — Begin with a simple workflow to verify results before building complex multi-step pipelines
Test first — Use the “test on 3 rows” option before applying formulas to your entire dataset
Iterate — After seeing initial results, ask the assistant to adjust prompts, change models, or add processing steps
Use undo freely — All changes are reversible, so experiment without worry

Limitations

Message length — Each message can be up to 5,000 characters
Spreadsheet preview — The assistant sees headers and a sample of your data (first rows and columns). For large datasets, it works from this sample and applies formulas to all rows
Long conversations — The assistant automatically summarizes older messages to maintain quality. Very long conversations work well thanks to this automatic context management
Phone verification required — Your account must have a verified phone number to use the AI Assistant