Most agent stacks either pay per query or outsource web retrieval to cloud services. Wigolo flips that tradeoff: it gives an AI agent a durable, local web layer that performs multi-engine search, robust fetching, site crawling, structured extraction, persistent caching, similarity search, and cited synthesis — all designed to run next to your agent with no API keys and no metered bill.
What Sets It Apart
- Local-first, privacy-first architecture: cache, embeddings, models, and config live under ~/.wigolo so queries and vectors stay on your machine unless you opt into an external LLM.
- Complete tool surface for agents: ten tools (search, fetch, crawl, extract, cache, find_similar, research, agent, diff, watch) accessible via MCP, REST, CLI, or SDKs (TypeScript/Python).
- Explainable evidence and provenance: results include verbatim excerpts pinned to byte offsets, a decomposed evidence score, and engine-level telemetry so agents can cite and inspect what they relied on.
- Robust fetch ladder and learned routing: automated escalation from plain HTTP to headless browser on anti-bot challenges, per-domain learning, robots.txt and rate-limit respect.
- $0 per query operational model: ranking, embeddings and reranker run on-device; search adapters talk to public engines but reranking & vectors are local, avoiding per-query cloud bills.
Who It's For & Trade-offs
Great fit if you self-host agents, need private or offline-capable web retrieval for RAG or autonomous gather loops, or want deterministic, inspectable evidence alongside synthesis. Integrations target coding-agent workflows (MCP clients, LangChain, LlamaIndex, Vercel AI SDK) and developer flows (CLI, Docker, Homebrew, npm/PyPI).
Look elsewhere if you need a fully managed, highly distributed hosted crawler at scale, if you cannot allocate ~1.5 GB disk for on-device models and a browser engine, or if you require enterprise SLAs and a vendor-hosted LLM stack out of the box. Wigolo is public beta and maintained under AGPL-3.0; it trades convenience of a hosted service for local control, transparency, and zero per-query cost.
Where It Fits
Positionally it sits between cloud metered web-retrieval services and single-page scraping scripts: it provides production-grade crawling/fetching and RAG-friendly outputs while keeping data and compute local. Use it to give coding agents a durable web surface or to replace paid web retrieval adapters in privacy-sensitive pipelines.
