pi-search-multi

Unified web search extension for pi with 9 backend providers (all working). One web_search tool, auto-fallback between backends.

Installation

pi install npm:pi-search-multi

Usage

After installing, just ask naturally:

Search for recent AI agent frameworks.

What's the latest news on Llama 4?

Or call the tool directly via web_search — the agent picks the best configured backend automatically.

Supported Backends

#	Backend	Free Tier	API Key?	How to get key
1	DuckDuckGo	Unlimited (rate-limited)	No	Nothing needed
2	Marginalia Search	Unlimited (rate-limited)	No†	marginalia.nu
3	Tavily	1,000 calls/month	Yes	tavily.com
4	Serper (Google)	2,500 queries/month	Yes	serper.dev
5	Brave	2,000 queries/month	Yes	brave.com/search/api
6	Firecrawl	500 free credits	Yes	firecrawl.dev
7	Exa	10 QPS rate-limited	Yes	exa.ai
8	LangSearch	Genuinely free, no CC	Yes	langsearch.com
9	WebSearchAPI.ai	2,000 free credits	Yes	websearchapi.ai

† Marginalia Search uses public as a shared API key — no registration required, but subject to a shared rate limit.

Removed: Stract, UnSearch, BoardReader, EntireWeb, Search1API, FreeAPITools.dev — no longer viable (public API removed, requires payment, or endpoint not implemented).

Benchmark Results (2026-05-04)

All 9 backends confirmed working across 3 test queries. All backends returning results were scored for relevance quality (0-10).

Latest benchmark run: 2026-05-04T18:34 UTC. Full report in benchmark/benchmark-report.md.

How Quality is scored: Each result is evaluated for keyword relevance (query words matched in title/snippet), source diversity (penalty for generic search engines), and snippet completeness. The average per-result score is then normalized to a 0–10 scale. Time is shown for reference only — it is not a factor in the quality score.

🏆 Working Backends

Backend	Avg Time	Quality	Status
Tavily	356ms	3.7/10	✅ Best quality, rich content
DuckDuckGo	1158ms	3.5/10	✅ Reliable, no key needed
Serper	667ms	3.5/10	✅ Google results
Firecrawl	644ms	3.5/10	✅ Search + crawl + extract
Brave	460ms	3.5/10	✅ Fast (~1 req/s free)
Exa	137ms	3.2/10	✅ AI-native search
Marginalia Search	354ms	3.0/10	✅ Fastest no-key backend
LangSearch	1816ms	3.2/10	✅ 10 results/query, free tier
WebSearchAPI.ai	1323ms	3.5/10	✅ Google-powered, 2K free credits

Configuration

Configure backends globally (all projects) or per-project:

Global: ~/.pi/agent/extensions/search.json Project: .pi/search.json (project takes precedence)

{
  "defaultBackend": "auto",
  "backends": {
    "duckduckgo": { "enabled": true },
    "marginalia": { "enabled": true },
    "serper": { "enabled": true, "apiKey": "your-serper-key" },
    "tavily": { "enabled": true, "apiKey": "your-tavily-key" },
    "brave": { "enabled": true, "apiKey": "your-brave-key" },
    "exa": { "enabled": true, "apiKey": "your-exa-key" },
    "firecrawl": { "enabled": true, "apiKey": "your-firecrawl-key" },
    "langsearch": { "enabled": true, "apiKey": "your-langsearch-key" },
    "websearchapi": { "enabled": true, "apiKey": "your-websearchapi-key" }
  }
}

See search.json.example for a full template.

Or use the interactive setup:

/search-setup

Commands

Command	Description
/search-setup	Interactive prompt to configure API keys for any backend
/search-status	Show which backends are active and which have keys

How auto mode works

1. Tries each enabled backend in order from your config
2. If a backend fails (rate limit, auth error, etc.), moves to the next one
3. DuckDuckGo requires no API key and is always included as a safety net
4. Returns results from the first backend that succeeds
5. If all backends fail, reports the collected errors

Security

• API keys are stored in local config files only (~/.pi/agent/extensions/search.json or .pi/search.json), never sent to any third party besides the chosen backend
• DuckDuckGo queries are executed via temp-file Python scripts (no shell injection surface)
• All HTTP backends have a 30-second timeout to prevent hanging requests
• Error messages are sanitized — API response bodies are truncated and key-like patterns are redacted before being returned
• The .pi/ directory is in .gitignore — never commit API keys to version control

Testing

# Run the full benchmark against all backends
node benchmark/benchmark.mjs

# Quick test via curl with your configured key
curl -X POST "https://api.exa.ai/search" \
  -H "Content-Type: application/json" \
  -H "x-api-key: $KEY" \
  -d '{"query": "test", "numResults": 3, "contents": {"text": true}}'

Adding a new backend

Backends are just async functions that return { results: [{ title, url, snippet }] }. See extensions/search.ts for examples.

License

MIT

---