Skip to content

Releases: chuanjin-su/pi-web-access-minimal

Release list

v0.1.2

v0.1.2 Latest
Latest

Choose a tag to compare

View all tags
@chuanjin-su chuanjin-su released this 12 Sep 08:26
v0.1.2
e9c27bf

Full Changelog: v0.1.1...v0.1.2

Assets 2
Loading

v0.1.1

v0.1.1

Choose a tag to compare

View all tags
@chuanjin-su chuanjin-su released this 12 Sep 08:11
v0.1.1
5c7eb69

Full Changelog: v0.1.0...v0.1.1

Loading

v0.1.0

v0.1.0

Choose a tag to compare

View all tags
@chuanjin-su chuanjin-su released this 12 Sep 08:03
v0.1.0
c68a158

v0.1.0 — Initial release

Minimal web search + URL fetching for the Pi coding agent. A stripped-down alternative to pi-web-access that keeps only the two core tools.

Install

pi install git:github.com/chuanjin-su/pi-web-access-minimal
# or pin the release:
pi install git:github.com/chuanjin-su/pi-web-access-minimal@v0.1.0
# try without installing:
pi -e git:github.com/chuanjin-su/pi-web-access-minimal

Highlights

web_search

  • Three providers: OpenAI (Responses API / Codex), Brave Search API, and Exa (keyless MCP — works with no API key at all).
  • Provider is resolved automatically from config — the agent never picks one, keeping tool descriptions (and token usage) small.
  • Automatic fallback chain (openai → brave → exa) when the configured provider fails.
  • Multi-query support: { queries: [...] } runs up to 3 queries concurrently, each returning its own synthesized answer.
  • recencyFilter (day/week/month/year), domainFilter (prefix with - to exclude), numResults (1–20).

fetch_content

  • Fetches URLs and extracts readable content as markdown.
  • Modes: readable (default extraction), raw (exact HTTP body), answer (page-local Q&A with a model).
  • Supports direct images, GitHub repositories, and PDFs (local unpdf extraction).
  • Large content is paged via an offset parameter instead of a third retrieval tool — full content is cached in memory per session.

Configuration

All optional; provider auth falls back to environment variables. Credential values accept $ENV_VAR, !command, or $$literal prefixes. See the README for the full flag reference (proxy, SSRF/domain policy, authFetch browser-cookie profiles, PDF limits, …).

{
  "searchProvider": "exa"
}
  • exaApiKey / EXA_API_KEY — optional; removes Exa MCP rate limits
  • braveApiKey / BRAVE_API_KEY — Brave Search API
  • openaiApiKey / OPENAI_API_KEY — or a Codex subscription via /login

Differences from pi-web-access

Dropped: curator browser UI + summary review, source_check, get_search_content (replaced by the offset parameter), YouTube/local-video analysis, and the remaining search providers and third-party fetch backends (Parallel, TinyFish, Firecrawl, Jina Reader, Crawl4AI, Gemini web, …).

Kept near-verbatim: openai-search.ts, brave.ts, exa.ts (minus inline-content handling, replaced by offset paging), and the HTTP/GitHub/PDF extraction pipeline — so targeted upstream fixes can be ported easily.

Quality

  • 25 tests (npm test), typecheck clean (npx tsc).

License

MIT (© 2026 Chuanjin Su; provider/extraction code adapted from pi-web-access, MIT © Nico Bailon).

Full Changelog: https://github.com/chuanjin-su/pi-web-access-minimal/commits/v0.1.0

Loading