Skip to main content
No connection or setup required — reference them directly in any step using @Anthropic, @OpenAI, or @Gemini. Use them to analyse documents, extract structured data, classify records, generate content, and reason across complex inputs.

​
Available AI connectors

ConnectorModelBest for
@AnthropicClaudeLong documents, legal and financial reasoning, complex extraction
@OpenAIGPTClassification, generation, structured extraction
@GeminiGeminiLong-context tasks, multimodal inputs
@Perplexity—Web-grounded research, finding current information
@Exa—Neural search across web content
@Anthropic, @OpenAI, and @Gemini require no setup. @Perplexity and @Exa require an API key — add them from Settings → Connectors.

​
Using AI models in steps

Describe what you want the AI to do in plain language. Felix constructs the prompt and API call automatically. 
Use @Anthropic to review each contract from the previous step and
identify clauses that deviate from our standard terms.
Return a list of deviations with a short explanation for each.

Use @OpenAI to classify each application from the previous step as
low risk, medium risk, or high risk based on the stated income,
loan amount, and credit history. Return a confidence score for
each classification.

​
Working with documents

AI connectors work directly with PDFs and documents. You don’t need to extract text manually before passing a document to an AI step. 
Pass each uploaded PDF to @Anthropic. Extract the applicant name,
requested amount, stated income, and any disclosed liabilities.
Output as a structured table.

​
Web research

Use @Perplexity or @Exa when a step needs to find current information from the web. 
Use @Perplexity to find the current regulatory requirements for
CASP licence applications in the Czech Republic. Return a
structured summary with source links.

​
Controlling where AI is applied

Felix applies AI only at the steps that need it. Steps that fetch data, write to a spreadsheet, or send a notification run as structured code — no AI involved. This keeps costs predictable and execution reliable.

​
Tips

  • Be specific about what to extract or produce. Precise descriptions produce consistent outputs.
  • Use @Anthropic for long or complex documents. Claude handles long-context tasks well — contracts, regulatory filings, and multi-page reports.
  • Use @Perplexity or @Exa for research steps. When a step needs current information rather than processing a document.
  • Add a Human Checkpoint after AI steps that produce content going downstream — especially before anything is sent or published.