ACTION_ID: ai_web_navigator
NAME: Web Pilot
CATEGORY: AI
CREDITS: 6

Navigate the browser step-by-step using Floqer. Use this when a
research task requires clicking, filling forms, or paging through a
site behind multi-step navigation — anything beyond a single-shot web
search.

1. INPUTS

task (string, required)
  Task. Step-by-step plan for browser navigation.

2. OUTPUTS

final_result (string) — Final Result. Placeholder: "XYZ is a B2B
  company".
steps_taken (jsonArray) — Steps Taken.
time_taken_seconds (number) — Time Taken (seconds). Placeholder: 0.
run_id (string) — Run Id.

3. HOW TO CONFIGURE

Configure Action body:

{
  "inputs": {
    "task": "Open https://example.com, click the 'Pricing' link, and return the price of the Pro plan."
  }
}

Field-by-field:
  - task   Free-text step-by-step plan describing what the browser
           agent should do. `{{ref}}` tokens (e.g.
           `{{input.company_website}}`) resolve per row, so every row
           runs its own navigation.

4. KEY NOTES

- One run per row. The `task` text resolves with that row's
  `{{ref}}` values, the agent navigates, and the result lands on the
  row.
- The `final_result` output is free text. If you need a structured
  shape downstream, follow with `llm_models` (with `output_format`)
  or `formula` to extract typed fields.
- Slower and more expensive per row than `llm_web_agents` — only
  reach for this when actual browser interaction is required.

5. WHERE IT FITS IN A WORKFLOW

Pattern: account list -> ai_web_navigator (step-by-step web nav) ->
llm_models (extract / structure the final_result) -> downstream.

6. WHEN TO USE

Use ai_web_navigator for step-by-step browser tasks (filling forms,
navigating multi-page sites, scraping behind logins, paging through
results).

7. WHEN NOT TO USE

Need single-shot web research with citations (no clicking)
  -> action-detail/llm_web_agents.txt

Need to scrape a single known URL
  -> action-detail/scrape_web_page_using_firecrawl.txt

Last updated: 2026-04-30.