curllm

LLM-DSL Quick Reference

Import

from curllm_core.llm_dsl import (
    LLMSelectorGenerator,
    LLMElementFinder,
    GeneratedSelector,
    ElementMatch,
    generate_selector
)
from curllm_core.form_fill.js_scripts import (
    generate_field_concepts_with_llm,
    find_form_fields_with_llm
)

1. Generate CSS Selector

generator = LLMSelectorGenerator(llm=my_llm)

# Consent checkbox
result = await generator.generate_consent_selector(page)

# Any field by purpose
result = await generator.generate_field_selector(page, "email input")

# Check result
if result.confidence > 0.5:
    await page.click(result.selector)

2. Find Element by Purpose

finder = LLMElementFinder(llm=my_llm, page=page)

result = await finder.find_element("login button")

if result:
    await page.click(result.selector)

3. Find Form Fields

selectors = await find_form_fields_with_llm(page, llm=my_llm)

# Returns: {'email': '[data-curllm-target="email"]', ...}

await page.fill(selectors['email'], "test@example.com")
await page.fill(selectors['message'], "Hello")
await page.click(selectors['submit'])

4. Generate Field Keywords

concepts = await generate_field_concepts_with_llm(page, llm=my_llm)

# Returns: {'email': ['email', 'e-mail'], 'phone': ['telefon', 'tel'], ...}

5. URL Resolution

from curllm_core.url_resolution import UrlResolver
from curllm_core.url_resolution.goals import TaskGoal

resolver = UrlResolver(page=page, llm=my_llm)
result = await resolver.find_url_for_goal(TaskGoal.FIND_CONTACT_FORM)

if result.url:
    await page.goto(result.url)

Result Types

@dataclass
class GeneratedSelector:
    selector: str       # CSS selector
    purpose: str        # What element does
    confidence: float   # 0.0 - 1.0
    method: str         # 'llm', 'statistical', 'fallback'
    reasoning: str      # Why this selector

@dataclass
class ElementMatch:
    selector: str
    confidence: float
    reasoning: str
    element_type: str   # input, button, link, etc.
    attributes: Dict[str, str]

Semantic Concept Pattern

# ❌ Don't do this
if data_key in ['phone', 'tel']:
    ...

# ✅ Do this
phone_concepts = {'phone', 'tel', 'telefon', 'mobile', 'komórka'}
if data_key in phone_concepts:
    ...

Common Concept Groups

email_concepts = {'email', 'mail', 'e-mail', 'correo', 'poczta'}
phone_concepts = {'phone', 'tel', 'telefon', 'mobile', 'komórka'}
name_concepts = {'name', 'imię', 'nazwisko', 'fullname', 'nombre'}
message_concepts = {'message', 'wiadomość', 'treść', 'content'}

What’s Allowed

Type	Example	OK?
Element type	input[type="email"]	✅
Semantic concept	{'email', 'mail'}	✅
Service pattern	.g-recaptcha	✅
Site-specific	#my-site-login	❌
if/elif chain	if 'email' in x:	⚠️ Convert

This site is open source. Improve this page.