AI Providers & Configuration
Batcher.ai connects to multiple AI providers to power your formulas. Configurations define which AI model to use and how it behaves. A free built-in model ("default") is included — add your own API keys to access premium models.
How configurations work
Section titled “How configurations work”A configuration is a saved set of parameters that tells Batcher.ai how to connect to an AI provider. Each configuration has a name that you reference in your formulas:
=LLM(A2, "my-chatgpt")Here, "my-chatgpt" is the name of a configuration you created.
You can create as many configurations as needed — for example, one for ChatGPT with a creative temperature, another for Claude with a formal system prompt, and a third for Perplexity for fact-checked responses.
Managing configurations
Section titled “Managing configurations”Configuration list
Section titled “Configuration list”Navigate to the Configurations page from the main menu. This page shows all your configurations in a table view.
From here you can:
- View all existing configurations at a glance
- Create a new configuration
- Edit an existing configuration
- Delete a configuration
Pre-configured defaults are provided to help you get started.
Creating a new configuration
Section titled “Creating a new configuration”Click ”+ Add New Configuration” and fill in the form:
- Name — Choose a descriptive name (this is what you use in formulas)
- Provider — Select the AI provider (OpenAI, Anthropic, Google, etc.)
- Model — Choose the specific model (e.g., gpt-4o, claude-sonnet-4-6)
- API Key — Enter your API key for the selected provider
- Adjust parameters — Fine-tune temperature, max tokens, system prompt, etc.
- Click “Create Configuration”
Supported AI providers
Section titled “Supported AI providers”| Provider | Models | Use case |
|---|---|---|
| Built-in (default) qwen3.5 27b instruct | Included open-source model | General purpose, free |
| OpenAI | GPT-5.4, GPT-5.2… | High-quality content generation |
| Anthropic | Claude Opus, Sonnet, Haiku | Analysis, writing, reasoning |
| Gemini 3.1 Pro… | Multi-modal, fast processing | |
| Ollama | your own instances | local inference, private IA infra |
| Vllm | your own instances | local inference, private IA infra |
| Perplexity | All models | Search-augmented, fact-checked responses |
| DeepSeek | DeepSeek Chat, DeepSeek Reasoner | Cost-effective, reasoning tasks |
| Mistral | Mistral Large, Medium, Small, Codestral | European AI, code generation |
| Groq | Llama 3.3, Gemma2 | Ultra-fast inference |
| Together AI | Llama, Qwen, Kimi | Open-source models at scale |
| Jina | Web reader | Web scraping — converts URL to markdown |
| Haloscan | SEO API | Keyword research, SERP analysis |
Additional providers are available. Check the provider dropdown in the configuration form for the full list.
Configuration parameters
Section titled “Configuration parameters”| Parameter | Description | Default | Example |
|---|---|---|---|
| Name | Identifier used in formulas | — | "my-chatgpt" |
| Model | The specific AI model to use | Depends on provider | gpt-5.4, claude-sonnet-4-6 |
| API Key | Your authentication key for the provider | — | sk-... |
| Max Tokens | Maximum length of the AI response in tokens | 2048 | 4096 |
| Temperature | Controls creativity: 0 = focused and deterministic, 1 = creative and varied | Provider default | 0.7 |
| Top P | Alternative to temperature — controls diversity by limiting token probability mass | Provider default | 0.9 |
| System Prompt | A fixed instruction prepended to every request using this configuration | None | "Always respond in English" |
| Frequency Penalty | Reduces repetition of words already used | 0 | 0.5 |
| Presence Penalty | Encourages the model to talk about new topics | 0 | 0.3 |
| Response Format | Expected format of the response | text | json |
| Vision | Enable image analysis for this configuration | Disabled | Enabled |
| Thinking | Show reasoning/thinking tokens in the response | Disabled | Enabled |
Special integrations
Section titled “Special integrations”Jina — Web scraping
Section titled “Jina — Web scraping”The Jina integration converts any URL into clean markdown content. Use it as the first step in a web data extraction workflow.
A2: https://example.com/product-pageB2: =LLM(A2, "jina")C2: =LLM(CONCAT("Extract the product name and price from: ", B2), "default")Cell B2 will contain the full page content in markdown format. Cell C2 processes that content with AI.
Haloscan — SEO keyword research
Section titled “Haloscan — SEO keyword research”Haloscan provides SEO data including keyword search volumes, cost-per-click, similar keywords, and SERP results. It returns JSON data that you can process with another AI call.
A2: wireless headphonesB2: =LLM(A2, "haloscan")C2: =LLM(CONCAT("Extract the top 10 keywords by search volume from: ", B2), "default")Haloscan also offers related keyword and question keyword modes for deeper SEO research.
- Name configurations descriptively — Use names like
"gpt4-creative"or"claude-formal"so formulas are self-documenting - Use system prompts — Set a system prompt for consistent behavior (e.g., “Always respond in bullet points” or “Always respond in French”)
- Adjust temperature for your use case — Low temperature (0.1-0.3) for factual extraction, high temperature (0.7-0.9) for creative content
- Start with the default model — Test your workflow with the free built-in model before switching to premium providers