# CrawlLama

A local AI research agent that answers questions through Ollama by combining web search, open-source intelligence lookups and multi-hop retrieval-augmented reasoning.

CrawlLama is a locally run AI research agent with an OSINT module for email, phone and IP intelligence, social media analysis and advanced search operators, combined with LangGraph-based multi-hop reasoning for complex queries: a router, search, analyze, follow-up, synthesize and critique workflow. An adaptive agent-hopping system selects a low, medium or high complexity agent automatically for a given query, escalating based on confidence, and an adaptive intelligence report can be toggled from the interactive settings menu.

It searches across multiple sources, including DuckDuckGo, Brave Search and the Serper API with fallback between them, plus a dedicated German and English Wikipedia search, and its retrieval-augmented generation system supports batch processing and multi-query hybrid search. Other features include a REST API built with FastAPI for integration, a plugin system for extensibility, an intelligent time-to-live cache with LRU eviction, and thread-pool parallelization for multi-aspect searches.

Local processing requires a local large-language-model backend such as Ollama; the project notes that cloud-based APIs such as Claude or OpenAI are not local. It is written in Python.

Website: https://github.com/arn-c0de/Crawllama