@demigodmode/pi-web-agent is a Pi package for web access.
Most agent web tools blur search, fetch, browser rendering, and research into one vague thing. pi-web-agent exposes one public research tool, web_explore, and keeps search/fetch/headless work inside that bounded workflow.
The point is keeping the model-facing boundary simple: ask web_explore to research a question, and it handles direct links, discovery, HTTP reads, targeted browser rendering, source ranking, source-quality checks, and caveats internally.
That sounds obvious, but a lot of agent tooling gets fuzzy right there. This package is meant to be stricter about what it actually did and more willing to say when a read was not good enough to trust. Bot-check pages, narrow source sets, unreadable threads, and cautionary/conflicting evidence should show up as caveats instead of fake confidence.
Compatibility notice: current pi-web-agent requires Pi 0.74+ because Pi packages moved to the @earendil-works/* scope. Update Pi before updating this package. If you are on an older Pi version, stay on @demigodmode/pi-web-agent@0.6.x until Pi is updated.
pi install npm:@demigodmode/pi-web-agentAfter installing, reload or restart Pi. Run /web-agent for the action menu, or /web-agent doctor to check whether the package loaded cleanly and which web backends are configured.
Headless rendering first tries a detectable Chromium-family browser: Chrome, Chromium, Edge, or Brave. If none is found, it falls back to Playwright-managed Chromium and still launches headless. Firefox/Safari-only systems can still use search and plain HTTP reads; browser-rendered fallback uses Chromium.
Later on, update installed packages with:
pi updateDocs site:
Work on the docs locally:
npm run docs:devBuild the docs:
npm run docs:buildpi-web-agent renders web tool output in one visible mode at a time:
compact— short summary, default everywherepreview— slightly richer bounded viewverbose— fuller bounded view
See the v0.3.0 release notes for a before/after of the transcript cleanup:
Primary UI:
/web-agent settings
Helper commands:
/web-agent doctor
/web-agent show
/web-agent changelog
/web-agent reset project
/web-agent reset global
/web-agent mode preview
/web-agent mode web_explore verbose
/web-agent mode web_explore inherit
Config files:
Global: ~/.pi/agent/extensions/pi-web-agent/config.json
Project: .pi/extensions/pi-web-agent/config.json
Precedence:
- built-in defaults
- global config
- project config
Project config overrides global config.
Example:
{
"presentation": {
"defaultMode": "compact",
"tools": {
"web_explore": { "mode": "verbose" }
}
}
}Backend config is also supported. Defaults remain DuckDuckGo search, plain HTTP fetch, and local-browser headless fallback with managed Chromium fallback configured.
Backend settings can be changed from:
/web-agent settings
Choose Backends to edit search/fetch providers, fallback behavior, and SearXNG or Firecrawl base URLs interactively. Firecrawl API keys should stay in environment variables rather than being written into config files.
If you already run SearXNG or Firecrawl, see the self-hosted backend guide:
npm install
npm test
npm run lint
npm run buildFor local Pi work, this repo includes .pi/extensions/pi-web-agent.ts.
If Pi is already running, use /reload after changes.
AGPL-3.0-only. See LICENSE.