A free, local alternative to Tavily
Search, extract, crawl, map and research for your agents — on your own machine, with no credits to count.
wigolo and Tavily, side by side
Tavily is a hosted search API for AI agents and RAG, with search, extract, crawl, map and research endpoints billed in credits.
wigolo is a free, open-source server that gives any AI agent the same web jobs from your own machine: search across 18 engines, fetch, crawl, extract, a persistent local cache, find-similar, research and autonomous gathering. There is no account and no metered bill.
| wigolo | Tavily | |
|---|---|---|
| Where it runs | Your machine, or your own server | Their cloud |
| Account / API key | None | Required |
| Cost per query | $0, unlimited | Metered (see pricing below) |
| Where your queries go | Straight to the search engines it queries and the pages it fetches — no vendor in between, no account attached | To the vendor, tied to your account |
| Persistent local memory | Every page lands in a local cache with keyword + semantic re-query, offline | — |
| Evidence format | Verbatim excerpts pinned to byte-offset source spans, with an explainable score per result | Results with snippets or page content |
| Agent surfaces | MCP (stdio + HTTP), REST, CLI, TypeScript & Python SDKs, LangChain, CrewAI, LlamaIndex, Vercel AI SDK | Hosted API, SDKs, MCP server |
| Pricing | Free | 1,000 free credits a month; plans from $30 for 4,000 credits; pay-as-you-go $0.008 per credit. Basic search is 1 credit, advanced 2 |
| Web search | 18 engines fused, reranked on-device, explainable scores | Search endpoint (basic / advanced depth) |
| Extract / crawl / map | fetch, extract, crawl (bfs, dfs, sitemap, map) | extract, crawl, map |
| Research | research: cited brief, LLM optional | Research endpoint |
Tavily details as published by Tavily, checked September 2026: Tavily credits & pricing. Check their site for current terms.
Moving from Tavily
Each Tavily endpoint has a wigolo tool that does the same job. Wire wigolo into your agent over MCP and it discovers these tools on its own; frameworks and scripts use the REST API or the TypeScript and Python SDKs.
| Tavily | wigolo tool | Notes |
|---|---|---|
search | search | search_depth from ultra-fast (cache only) to deep |
extract | fetch / extract | |
crawl | crawl | |
map | crawl | strategy: "map" |
research | research |
Full parameters for every tool: tool reference.
Questions
Can I swap Tavily out of a LangChain or CrewAI agent?
Yes. wigolo ships LangChain, CrewAI and LlamaIndex integrations and a Vercel AI SDK tool set, plus TypeScript and Python clients, so the web tools in your agent point at your own machine instead of a metered API.
How fast is it without a hosted index?
search_depth trades depth for speed: ultra-fast answers from the local cache, fast queries engines only, balanced (the default) adds enrichment, deep goes furthest. Repeat questions get cheaper as the cache warms.
Does wigolo need an API key?
No. Search, fetch, crawl, extract and the cache run keyless on your machine. An LLM key (or a local model) is optional and only adds answer synthesis to research and agent runs.
Which agents does it work with?
Any MCP client. npx wigolo init --agents=<agent> wires Claude Code, Cursor, Codex, Gemini CLI, OpenCode, VS Code, Windsurf, Zed and Antigravity in one command; frameworks and self-hosted automations use the REST API or the SDKs.
Is it really free?
Yes — wigolo is open source (AGPL-3.0) with no paid tier and no metered bill. It runs on hardware you already have, so there is nothing to charge per query. Companies that want to embed it without AGPL obligations can take a commercial license.