KnockOutEZ/wigolo

What it solves

wigolo provides a local-first, keyless web intelligence layer for AI agents, eliminating the need for paid cloud APIs, metered billing, and third-party data storage for common web tasks. It allows agents to search, crawl, and extract data from the web while keeping the cache, embeddings, and models on the user's local machine.

How it works

It operates as a single Node.js process that can be interfaced via the Model Context Protocol (MCP), a REST API, or dedicated SDKs. The system uses a tiered fetch router to handle everything from plain HTTP to headless browsers for complex sites. It integrates 18 search engine adapters with rank fusion and on-device ML reranking to provide explainable evidence scores for results. While core tools are keyless, users can optionally connect an LLM (like Gemini or Ollama) for synthesized research reports.

Who it’s for

• AI Agent Developers: Those building with frameworks like LangChain, CrewAI, or LlamaIndex who need a robust web toolset.
• Coding Agents: Users of Claude Code, Cursor, Windsurf, or Zed who want to extend their agent's web capabilities locally.
• Self-Hosters: Developers who prefer private, local-first infrastructure over cloud-based search and scraping services.

Highlights

• Keyless & Free: Core search, fetch, and crawl tools require no API keys and cost $0 per query.
• Local-First Privacy: Cache, embeddings, and on-device models are stored locally in ~/.wigolo/.
• Comprehensive Toolset: Includes 10 specialized tools for searching, fetching, crawling, structured extraction, semantic caching, and autonomous research loops.
• Agent-Optimized Output: Provides verbatim excerpts pinned to byte-offset source spans and explainable score decompositions for high-fidelity evidence.
• Broad Integration: Supports MCP, REST, TypeScript/Python SDKs, and major agent frameworks.