Verity: Local AI Search Engine Like Perplexity

What It Is

Verity is an open-source search tool that brings AI-powered answer generation to local machines without relying on cloud services. Unlike commercial options that process queries on remote servers, Verity runs entirely on-device, combining web search results with local language model inference to generate comprehensive answers.

The architecture connects to SearXNG, a privacy-focused metasearch engine that aggregates results from multiple sources, then feeds those results to a local language model for synthesis. By default, Verity uses the Jan-nano 4B model, though developers can swap in any compatible model that fits their hardware constraints. The system leverages OpenVINO for Intel AI PCs or Ollama for broader compatibility, with acceleration support across CPU, GPU, and NPU hardware.

Two interaction modes accommodate different workflows: a command-line interface for quick lookups and a web-based interface for more exploratory research sessions. This flexibility makes Verity practical for both scripted automation and interactive browsing.

Why It Matters

Privacy-conscious developers and organizations gain a viable alternative to cloud-based AI search without sacrificing functionality. Every query, search result, and generated answer stays on the local machine, eliminating concerns about data collection, query logging, or third-party access to sensitive research topics.

The timing aligns with growing interest in local-first software and increasing hardware capabilities. Modern processors, particularly Intel’s Core Ultra series with dedicated NPU units, now handle language model inference at speeds that make local AI search practical rather than theoretical. Verity capitalizes on this hardware evolution by distributing workload across available accelerators.

For enterprise environments with strict data governance requirements, Verity offers a path to AI-enhanced search without introducing external dependencies or compliance risks. Research teams working with confidential information can explore topics freely without creating audit trails on third-party platforms.

The open-source nature also matters for customization. Teams can modify search sources, adjust model selection based on accuracy versus speed tradeoffs, or integrate Verity into existing local-first toolchains. This extensibility distinguishes it from closed commercial alternatives.

Getting Started

Installation requires cloning the repository and setting up dependencies:

Before running Verity, developers need a SearXNG instance accessible locally or on the network. The project repository at https://github.com/rupeshs/verity includes configuration details for connecting to search backends.

For command-line searches, the basic syntax processes queries and returns synthesized answers directly in the terminal. The web interface launches a local server that presents a familiar search box and displays both source results and AI-generated summaries.

Hardware acceleration configuration varies by platform. Intel AI PC users benefit from automatic NPU detection through OpenVINO, while other systems default to CPU inference or can configure GPU acceleration if available. Model selection happens through configuration files, allowing teams to balance response quality against inference speed based on available resources.

Context

Verity occupies a niche between fully local search tools and cloud-based AI assistants. Traditional local search solutions like Recoll or DocFetcher index files but lack natural language understanding. Cloud services like Perplexity or Bing Chat provide sophisticated AI synthesis but require internet connectivity and data sharing.

The main limitation is model size constraints. While 4B parameter models run efficiently on consumer hardware, they produce less nuanced answers than the larger models powering commercial services. Teams needing state-of-the-art reasoning might find local models insufficient for complex queries.

SearXNG dependency adds setup complexity compared to standalone tools. Organizations without existing SearXNG deployments face additional configuration overhead, though the privacy benefits often justify the effort.

Alternatives include running Ollama with custom search integrations or using RAG frameworks like LangChain with local embeddings. These approaches offer more flexibility but require more development work. Verity provides a ready-made solution for teams wanting AI search without building custom infrastructure.

The project suits developers comfortable with self-hosting and organizations prioritizing data sovereignty over cutting-edge model performance.