wigolo: a local-first web intelligence layer for AI agents with keyless search and byte-pinned evidence

What it solves

wigolo provides a local-first web intelligence layer for AI agents, eliminating the need for expensive, metered cloud APIs for web search, crawling, and data extraction. It allows agents to access the live web privately and for free, while providing high-quality, verifiable evidence (byte-pinned excerpts) that prevents hallucinations.

How it works

It operates as an MCP (Model Context Protocol) server, REST API, or SDK that integrates with coding agents (like Cursor or Claude Code) and agent frameworks. The system uses a tiered fetch router that escalates from plain HTTP to headless browsers to bypass bot challenges. It performs multi-engine search with rank fusion and on-device ML reranking, storing all cache and embeddings locally in ~/.wigolo/ to ensure privacy and zero cost per query.

Who it’s for

• AI Agent Developers: Those building with LangChain, CrewAI, LlamaIndex, or Vercel AI SDK who need a robust web toolset.
• Power Users of AI Editors: Users of Cursor, Windsurf, Zed, or VS Code who want their coding agents to have better, private web access.
• Self-Hosters: Developers who want a completely local web-intelligence stack without API keys or monthly bills.

Highlights

• Keyless Operation: Search, fetch, crawl, and extraction work without any API keys.
• Verifiable Evidence: Returns verbatim excerpts pinned to exact byte-offsets in the source for high provenance.
• Comprehensive Toolset: Includes search, multi-page crawling, structured data extraction, semantic caching, and autonomous research loops.
• Local-First Privacy: All data, embeddings, and models are stored on the user's machine.
• Broad Integration: Supports MCP, REST, TypeScript/Python SDKs, and major agent frameworks.