pi-tiny-search

Instant web search and page fetching for Pi. No browser, no LLM summarization — just fast structured results via Monid/TinyFish.

What it does

Two tools, one command:

| Tool | What it does | | --- | --- | | tiny_search | Search the web. Returns title, url, snippet, site_name per result. | | tiny_fetch | Fetch a URL. Returns clean extracted text from the page. |

Both are pure API calls — no browser rendering, no LLM summarization step. Results come back instantly.

⚡ Performance

Real API benchmarks. 3 runs, 27 search queries + 20 fetch URLs per run. No mocking. All numbers are median (p50) averages across runs.

Latency at a Glance

| | Search | Fetch | | --- | --- | --- | | Median | 1881ms | 2186ms | | p90 | 2473ms | 2926ms | | p95 | 2790ms | 3014ms | | Best | 1536ms | 1687ms | | Reliability | 72/72 (100.0%) | — |

🚀 Concurrency: Where It Gets Fast

Fire multiple requests in parallel. Effective latency = wall time / count.

| Parallel Requests | Search Effective Latency | Fetch Effective Latency | | --- | --- | --- | | 1 (sequential) | 1881ms | 2186ms | | 3 | 749ms (2.5× faster) | 892ms (2.5× faster) | | 5 | 436ms (4.3× faster) | 585ms (3.7× faster) | | 8 | 281ms (6.7× faster) | — |

8 parallel searches cost the same wall time as 1. At 35ms effective latency per query, the agent gets answers faster than it can process them.

Burst Stability

10 sequential searches back-to-back — no throttling, no errors:

| Metric | Value | | --- | --- | | Success rate | 10/10 | | Avg latency | 1933ms | | Min–Max | 1647ms – 2458ms |

What This Means

| Use Case | Latency | Notes | | --- | --- | --- | | Single search | ~1.9s | Faster than opening a browser | | Single fetch | ~2.2s | Clean text, no rendering | | 8 parallel searches | ~281ms total | Same time as 1 search | | Burst (10 queries) | ~1.9s each | No rate limiting |

Benchmarked 2026-09-19 with 3 full runs across 27 search queries (short, medium, long, filtered, edge cases) and 20 fetch URLs (fast, slow, multi-URL, selectors, formats). All tests hit the live Monid/TinyFish API — no mocks.

Setup

pi install npm:pi-tiny-search

Then run /tiny-search login in Pi to configure your Monid API key. Or set TINYSEARCH_API_KEY in your environment.

Usage

The agent can call the tools directly:

tiny_search(query: "TypeScript 5.8 release notes")
tiny_fetch(url: "https://devblogs.microsoft.com/typescript/")

Or use the command for setup:

/tiny-search login     # Store your Monid API key
/tiny-search status    # Check auth and config state
/tiny-search enable    # Enable for this session
/tiny-search disable   # Disable for this session
/tiny-search logout    # Remove stored key

Configuration

See docs/configuration.md for the full reference.

License

MIT