pi-querit

August 27, 2026 · View on GitHub

A focused Pi extension that gives your agent live web search and page fetching through Querit.

Why Querit?

LLMs are limited by their training data — complex or real-time queries lead to hallucinations and stale answers. Querit is a retrieval system built specifically for generative LLM invocation scenarios, delivering real-time, authoritative web search results that integrate directly into LLM applications:

Comprehensive content — a massive global index spanning nearly 20 countries and 10 languages with hundreds of billions of web pages.
Strong capabilities — flexible retrieval options (time range, region, language, and domain filters) so results can be tuned for specific scenarios.
Excellent results — accurate, authoritative, high-quality content coverage.

Quick Start

1. Install

pi install npm:pi-querit

2. Configure

Start Pi interactively and run the setup wizard:

/querit-setup

The wizard walks you through API key entry and search defaults. Every step after the API key can be skipped.

3. Search

Just ask your agent anything that needs the web — it calls web_search and fetch_content automatically:

> What changed in the Rust 2024 edition?

Note: Do not enable this package together with another extension that also registers web_search or fetch_content; Pi tool names must be unique.

How It Works

┌──────────────────────────────────────────────────────────┐
│  Pi Agent                                                │
│                                                          │
│  web_search(query, count?)                               │
│      │                                                   │
│      ▼                                                   │
│  ┌─────────────┐    persistent defaults     ┌─────────┐  │
│  │  Querit API  │◄── (timeRange, countries, │ querit-  │  │
│  │  /v1/search  │    languages, domains…)   │ search.  │  │
│  └──────┬──────┘    from /querit-setup      │ json     │  │
│         │                                   └─────────┘  │
│         ▼                                                │
│    return cited results to the model                     │
│                                                          │
│  fetch_content(url/urls, format?)                        │
│      │                                                   │
│      ▼                                                   │
│  ┌──────────────────┐                                    │
│  │  Querit API       │                                   │
│  │  /v1/contents     │──► markdown / text / HTML         │
│  └──────────────────┘    (up to 10 URLs per call)        │
└──────────────────────────────────────────────────────────┘

The extension registers two tools and one slash command:

Surface	Purpose
`web_search`	Live web search with cited results
`fetch_content`	Full page content (markdown, text, or HTML) for up to 10 URLs per call
`/querit-setup`	Interactive configuration of the API key and persistent search defaults

Configuration

Run /querit-setup in Pi's interactive mode. Configuration is stored in ~/.pi/agent/querit-search.json (respects PI_CODING_AGENT_DIR).

First-time setup flow

API key — masked input prompt; validated with a one-result search request.
Search defaults — each step can be skipped (see table below).

When a key is already configured, /querit-setup opens a menu:

Option	Effect
Replace API key (full re-setup)	Prompts for a new key, validates it, re-runs the full flow. The old key is overwritten locally; revoke it in the Querit dashboard if needed.
Change search defaults	Edits the persistent search filters without touching the saved key.

Search default options

These are persistent defaults applied to every web_search call. They are stored under the search key in querit-search.json and are not per-call parameters.

Option	Values	Default	Description
`count`	`1` – `20`	API default (`5`)	Number of results per search. Can be overridden per call via the `count` parameter.
`timeRange`	`d7` · `w2` · `m3` · `y1`	(none — all time)	Restrict results to the past 7 days, 2 weeks, 3 months, or 1 year.
`includeContent`	`yes` / `no`	`no`	Include sentence-level content excerpts in results for richer context.
`chunksPerDoc`	`1` – `3`	API default (`1`)	Content chunks requested per result. This advanced option is configured by editing `querit-search.json`; it is not exposed in the setup wizard.
`countries`	`argentina` · `australia` · `brazil` · `canada` · `colombia` · `france` · `germany` · `india` · `indonesia` · `japan` · `mexico` · `nigeria` · `philippines` · `south korea` · `spain` · `united kingdom` · `united states`	(none — global)	Bias results toward specific countries. Comma-separated, multi-select.
`languages`	`english` · `japanese` · `korean` · `german` · `french` · `spanish` · `portuguese`	(none — all)	Filter results by language. Comma-separated, multi-select.
`includeDomains`	domain list	(none — unrestricted)	Whitelist — only these domains return results.
`excludeDomains`	domain list	(none)	Blacklist — these domains are excluded. Ships with a built-in Noise blockers preset: `pinterest.com`, `facebook.com`, `instagram.com`, `tiktok.com`.

Configuration file example

{
  "apiKey": "your-api-key",
  "search": {
    "count": 5,
    "chunksPerDoc": 1,
    "timeRange": "m3",
    "includeContent": false,
    "countries": ["united states"],
    "languages": ["english"],
    "includeDomains": ["github.com"],
    "excludeDomains": ["pinterest.com"]
  }
}

The extension applies mode 0600 on POSIX systems. On Windows, the file remains protected by the user profile's filesystem ACLs. The key is never included in tool results or logs.

For CI or ephemeral use, set QUERIT_API_KEY. The environment variable takes precedence when both are present.

Tools

`web_search`

Required:

query

Optional:

count (1..20) — overrides the configured default for one call (API default: 5)

Domains, time range, countries, languages, and content excerpts are persistent defaults configured in /querit-setup (stored under search in querit-search.json), not per-call parameters. Skipping the domain lists leaves search unrestricted; the include list is a whitelist (only those domains return results), the exclude list is a blacklist.

Results include explicit title, URL, snippet, source metadata, and optional sentence excerpts. Duplicate and non-HTTP(S) result URLs are removed.

`fetch_content`

Pass url, urls, or both (at most 10 unique HTTP(S) URLs).

Optional:

format: markdown (default), text, or html
crawl_timeout: 1..60 seconds (default 10)
include_metadata: default true

Both tools mark remote data as untrusted, propagate Pi cancellation, enforce response-size limits, and cap model-visible output at Pi's 50KB/2000-line limit. If formatted output is truncated, the complete output is written to a uniquely named file inside a single shared temp folder (pi-querit in the OS temp directory). To keep search content private, that folder never accumulates files: the files a Pi process creates are deleted when that session shuts down (with a process-exit fallback), and leftovers from crashed runs are swept the next time the extension loads.

Development

Requires Node.js 22.19 or newer.

npm install
npm run check
npm test
npm run pack:check

A live smoke test reads the same JSON configuration (or QUERIT_API_KEY) and exercises both APIs without printing the key or fetched content:

npm run test:live

License

MIT