realtimex-lead-search 0.2.0

realtimex-lead-search

Lead discovery and qualification flows for RealtimeX. Local-first scraping and storage; user-selected LLM (cloud or local). No auto-switching: cloud failures surface to the user.

Structure

• realtimex_lead_search/lead_search_agent.py — entry point (run via uv run -m realtimex_lead_search.lead_search_agent).
• realtimex_lead_search/lead_search/ — modules for scraping, LLM, extraction, scoring, data, cache, strategies, anti-detection, models, prompts.
• docs/realtimex-lead-search-module-io.md — module IO reference.
• tests/ — pytest coverage for strategies, extraction/scoring, and persistence.

Getting started

1. Use uv run -m realtimex_lead_search.lead_search_agent --payload payload.json to execute (or pipe JSON to stdin). No pip install inside RealtimeX flows.
2. Payload shape (extra fields ignored):

{
  "keywords": ["plumber near me"],
  "locations": ["seattle"],
  "sources": ["google_maps"],
  "pages_per_source": 2,
  "filters": {"categories": ["plumbing"], "must_have_phone": true},
  "llm": {"provider": "openai", "model": "gpt-4.1-mini", "base_url": "https://api.openai.com", "api_key": "sk-..."},
  "storage": {"sqlite_path": "./data/lead_search.db", "json_export": true}
}

3. Optional: use_llm_extraction: true or CLI --use-llm to enable LLM parsing in addition to heuristics.
4. Keep LLM selection user-driven; do not auto-fallback between providers.

Notes

• Keep everything local except the user-selected LLM call.
• Respect robots/delays; skip captchas rather than solving.