Wigolo

Local-first web intelligence for AI agents — no keys, no cloud, no metered bill.

works with Claude Code · Cursor · Codex · Gemini CLI · OpenCode · VS Code · Windsurf · Zed · Antigravity

and beyond LangChain · CrewAI · LlamaIndex · Vercel AI SDK · n8n & self-hosted agents · any MCP client · plain REST

sponsored by

Quickstart · Tools · Why wigolo · Discord · Sponsors · Benchmark · Docs · Examples · Licensing · Feedback · FAQ

Join the community on [Discord](https://discord.gg/BkUUgz2bNF) — questions, help, and what's being built next.

New features and updates ship steadily. Follow @yourtowhid on X for all of it and new ways to use wigolo, and reach out there for collaborations or feedback · also on LinkedIn

---

wigolo gives an AI agent one surface for everything web-related: search, fetch, crawl, extract, cache, find-similar, research, and autonomous gather loops. It runs wherever your agent runs — as an MCP server next to your coding agent, as a REST/MCP endpoint on the box where your self-hosted agents live, or embedded through an SDK inside your own app. The core tools need no API keys, nothing it touches leaves ~/.wigolo/, and no bill grows with how much your agent thinks.

Quickstart

npx wigolo init                              # set up the local engine — any system

npx wigolo init --agents=claude-code,cursor # …or set up + wire your day-to-day agents in one command

Requires Node ≥ 20 and ~1.5 GB of free disk on macOS, Linux, or Windows. Bare init sets up the local engine: it downloads the browser engine and on-device models, runs a health check, and reports each component. Adding --agents wires the named agents in the same run, so a coding agent you use daily is ready in one command.

• Supported agents — --agents takes any of claude-code · cursor · codex · gemini-cli · opencode · vscode · windsurf · zed · antigravity (comma-separated); wigolo writes the MCP config and, where supported, instructions for each.
• Any other setup — any MCP client, agent framework, or self-hosted agent registers npx -y wigolo in its own MCP config. The installation guide has the exact config block for every client, plus Docker, Homebrew, and single-file-binary channels.
• More on the way — the supported list keeps growing, and a PR to add your agent is welcome; see CONTRIBUTING.md.
• Interactive setup — --interactive is a plain-text flow; --wizard is the full terminal TUI.
• Defer downloads — --no-warmup waits until first use. A failed component download never fails setup; init reports what's not ready with the exact fix and still completes.

init is unattended by default, so it's safe in scripts and CI, and any setup problem surfaces right here in the per-component report, before your agent's first call. Search, fetch, crawl, extract, cache, and find-similar work with no API key. Check it's healthy anytime:

npx wigolo doctor

To remove everything cleanly, run npx wigolo config --uninstall --yes. You can also paste the installation guide into any AI assistant and let it do the setup; it's written to be self-contained.

Recommended — a free key for research & agent

Search, fetch, crawl, extract, cache, and find-similar are fully keyless. research, agent, and search format=answer use an LLM to write the synthesized, cited answer. Without one they hand back a raw brief and evidence for your agent to assemble. A free Gemini key turns that into a finished answer:

export WIGOLO_LLM_PROVIDER=gemini

export GEMINI_API_KEY= # grab one at aistudio.google.com/apikey — the free tier is plenty

Any provider works (anthropic · openai · groq), or stay fully local and keyless with WIGOLO_LLM_PROVIDER=ollama (or any OpenAI-compatible URL). Set it in your shell or your agent's MCP env block. Providers, models, and the keyless local-model ladder are in the configuration guide.

What your agent gets back

Every search result is evidence the agent can act on. It carries a verbatim excerpt pinned to its exact position in the source, a citation ID the agent can quote, and a score it can inspect (abridged real shape):

{

"results": [{ "title": "Logical replication - PostgreSQL docs", "url": "https://www.postgresql.org/docs/current/logical-replication.html", "excerpt": "Logical replication is a method of replicating data objects…", "citation_id": "src-1", "source_span": { "start": 1042, "end": 1305 }, // byte-exact provenance "evidence_score": { "final": 0.86, "semantic": 0.91, "lexical": 0.78, "engine_consensus": 3 } }], "citations": [{ "id": "src-1", "url": "…" }], "freshness_signal": { "published": "2026-05-12", "confidence": "high" } }

Weak results get flagged as junk by wigolo's own scorer. Failed engines are reported and stale cache is labeled, so the agent always knows what it's standing on. Full response contracts per tool are in the tools reference.

Tools

Every tool also runs from the terminal (wigolo search "…" --json), from an interactive shell with NDJSON piping (wigolo shell), over REST, and through the SDKs — CLI reference. Per-tool guides with the full parameter set are in docs/tools.md; runnable examples are in examples/.

Why it's different

wigolo isn't a free stand-in for the paid tools — it's built to match them. It's a focused web layer for your agents: an MCP and REST surface they call directly, with the search and extraction quality the paid services charge for. What separates it:

• Built for agents. One MCP call fans out many queries across many engines in parallel, which a serial host tool-loop can't replicate. Every result carries transparent per-result scoring, and output is budget-aware.
• Honest output. Stale cache, failed fetches, degraded backends, and truncation are surfaced in the result. When a bot-protected page can't be read, you get a labeled blocked_by_challenge failure, not a challenge shell returned as content.
• $0 per query, free to re-query. Default search talks to public engines through direct adapters; the reranker and embeddings run on-device. Every response is cached, so asking again is instant and costs nothing.
• Private by default. Cache, embeddings, models, and config live under ~/.wigolo/. Nothing reaches a third party unless you explicitly opt into an LLM for synthesis.

Here's what one real result looks like, dissected. It includes the failed engine and the weak result, because those are part of the answer too:

Benchmark

All four tools converged on the same core answer, and only one of them handed back verbatim, byte-pinned evidence while doing it.

One cold query ran live inside a single Claude Fable 5 session, fanned out to four web tools on equal footing (built-in WebSearch, wigolo, Tavily, Exa), and was judged by the agent on the evidence alone. All four converged on the same answer and the same top source, so the parity is demonstrated on-screen. wigolo alone returned verbatim excerpts pinned to byte-offset source spans, an explainable score d