Descriptions:
Fahd Mirza walks through a complete installation and demonstration of Local Deep Research, an open-source AI research assistant that runs entirely on a local machine without any API keys. Using Docker Compose on Ubuntu with an Nvidia RTX A6000 GPU, Mirza spins up three containers — Ollama for local LLM inference, SearXNG as a self-hosted meta search engine, and the Local Deep Research application itself — and shows exactly how they interconnect.
The tool works by decomposing a research query into sub-questions, then searching in parallel across the live web (via SearXNG), academic databases including arXiv and PubMed, and locally stored documents. Every source retrieved is downloaded to an encrypted local library and indexed as vector embeddings, enabling semantic search over previously collected material. The knowledge base compounds over time: the more queries run, the more future searches draw on local data rather than external services.
Mirza also captures a real-world troubleshooting scenario — a permission bug in the SearXNG Docker container that prevents it from starting — and resolves it live. The demo uses Gemma 7B via Ollama, though any Ollama-compatible model can be substituted. The interface at localhost:5000 supports detailed reports or quick summaries and offers configuration options for embedding models, cosine similarity thresholds for semantic search, and alternative model providers including llama.cpp or hosted APIs.
📺 Source: Fahd Mirza · Published May 08, 2026
🏷️ Format: Tutorial Demo
