# Pi Web Access
**Web search, content extraction, and video understanding for Pi agent. Zero-config Exa search, optional browser-cookie Gemini Web, or bring your own API keys.**
[](https://www.npmjs.com/package/pi-web-access)
[](https://opensource.org/licenses/MIT)
[]()
https://github.com/user-attachments/assets/cac6a17a-1eeb-4dde-9818-cdf85d8ea98f
## Why Pi Web Access
**Zero Config** — Works out of the box with Exa MCP (no API key needed). Add API keys for Exa, Perplexity, or Gemini API for more control, or opt into browser-cookie access for Gemini Web.
**Video Understanding** — Point it at a YouTube video or local screen recording and ask questions about what's on screen. Full transcripts, visual descriptions, and frame extraction at exact timestamps.
**Smart Fallbacks** — Web search tries Exa, then Perplexity, then Gemini API, then Gemini Web when browser cookies are enabled. YouTube tries Gemini Web when enabled, then API, then Perplexity. Blocked pages retry through Jina Reader and Gemini extraction. Repository-specific `code_search` uses DeepWiki only.
**GitHub Cloning** — GitHub URLs are cloned locally instead of scraped. The agent gets real file contents and a local path to explore, not rendered HTML.
## Install
```bash
pi install npm:pi-web-access
```
Works immediately with no API keys — Exa MCP provides zero-config search. For more providers or direct API access, add keys to `~/.pi/web-search.json`:
```json
{
"exaApiKey": "exa-...",
"perplexityApiKey": "pplx-...",
"geminiApiKey": "AIza..."
}
```
In `auto` mode (default), `web_search` tries Exa first (direct API if keyed, MCP if not), then Perplexity, then Gemini API, then Gemini Web when browser-cookie access is enabled.
Optional dependencies for video frame extraction:
```bash
brew install ffmpeg # frame extraction, video thumbnails, local video duration
brew install yt-dlp # YouTube stream URLs for frame extraction
```
Without these, video content analysis (transcripts, visual descriptions via Gemini) still works. The binaries are only needed for extracting individual frames as images.
Requires Pi v0.37.3+.
## Quick Start
```typescript
// Search the web
web_search({ query: "TypeScript best practices 2025" })
// Fetch a page
fetch_content({ urls: ["https://docs.example.com/guide"] })
// Clone a GitHub repo
fetch_content({ urls: ["https://github.com/owner/repo"] })
// Understand a YouTube video
fetch_content({ urls: ["https://youtube.com/watch?v=abc"], prompt: "What libraries are shown?" })
// Analyze a screen recording
fetch_content({ urls: ["/path/to/recording.mp4"], prompt: "What error appears on screen?" })
```
## Tools
The public tools register immediately and share one lazy initialization attempt on first use. Calls wait for initialization and active-session replay before provider work begins; a rejected initializer or replay can be retried by a later call. Initialization and replay are leased to the active session: shutdown retires the lease and cleans its candidate, so calls spanning shutdown reject and a restarted session creates a fresh provider wrapper. An abort that arrives during the shared wait cancels only that invocation. Host cancellation after provider or curator work rejects with the exact host abort reason, while an explicit user action in the curator remains a successful `{ cancelled: true, cancelReason: "user" }` result. Resetting the application-owned attempt does not guarantee re-evaluation of an ESM module whose evaluation itself failed.
For batch search/fetch calls, partial success remains usable and retains item-level errors. A non-empty batch where every provider request or URL fetch fails is reported through Atomic's tool-error channel with aggregate stage metadata while the per-item diagnostics remain stored in the result.
### web_search
Search the web via Exa, Perplexity AI, or Gemini. Returns a synthesized answer with source citations. The curator UI is off by default; enable it with `/curator on` or `"workflow": "summary-review"` in config.
```typescript
web_search({ query: "rust async programming" })
web_search({ queries: ["query 1", "query 2"] })
web_search({ query: "latest news", numResults: 10, recencyFilter: "week" })
web_search({ query: "...", domainFilter: ["github.com"] })
web_search({ query: "...", provider: "exa" })
web_search({ query: "...", includeContent: true })
```
| Parameter | Description |
|-----------|-------------|
| `query` / `queries` | Single query or batch of queries |
| `numResults` | Results per query (default: 5, max: 20) |
| `recencyFilter` | `day`, `week`, `month`, or `year` |
| `domainFilter` | Limit to domains (prefix with `-` to exclude) |
| `provider` | `auto` (default), `exa`, `perplexity`, or `gemini` |
| `includeContent` | Fetch full page content from sources in background |
### code_search
Ask questions about code, architecture, and APIs in a public GitHub repository via DeepWiki MCP at `https://mcp.deepwiki.com/mcp`. No API key or local MCP configuration is required. DeepWiki availability and repository indexing determine which questions it can answer.
```typescript
code_search({ repoName: "facebook/react", query: "How does useEffect cleanup work?" })
code_search({ repoName: "expressjs/express", query: "How is middleware error handling implemented?", maxTokens: 10000 })
```
| Parameter | Description |
|-----------|-------------|
| `repoName` | Required single public GitHub repository in `owner/repo` format, not a URL or list |
| `query` | Required nonempty question about that repository, sent verbatim |
| `maxTokens` | Optional best-effort output bound, approximately four characters per token, plus a truncation notice (default: 5000, range: 1000–50000) |
Migration: existing `code_search({ query: ... })` calls must now include `repoName`. The query is sent as DeepWiki's `ask_question` question. `maxTokens` limits the returned text locally, not DeepWiki's generation. Requests have a 60-second deadline and honor caller cancellation. Errors and empty responses are reported without falling back to Exa. Use `web_search` for broader discovery or when a repository is unavailable; its providers and settings are unchanged.
### fetch_content
Fetch URL(s) and extract readable content as markdown. Automatically detects and handles GitHub repos, YouTube videos, PDFs, local video files, and regular web pages.
Use the lowercase `urls` field with a nonempty array of strings, even for one URL. Replace legacy `{ url: "..." }` calls with `{ urls: ["..."] }`. Missing targets, empty arrays, empty strings, object entries, and unrecognized fields are rejected before fetching. Atomic's standard argument normalization can convert a scalar `urls` string into a one-item array, but prompts and integrations should always send the documented array form.
See [Atomic's fetch argument guide](../coding-agent/docs/web-access.md) for examples and validation troubleshooting.
```typescript
fetch_content({ urls: ["https://example.com/article"] })
fetch_content({ urls: ["https://example.com/one", "https://example.com/two"] })
fetch_content({ urls: ["https://github.com/owner/repo"] })
fetch_content({ urls: ["https://youtube.com/watch?v=abc"], prompt: "What libraries are shown?" })
fetch_content({ urls: ["/path/to/recording.mp4"], prompt: "What error appears on screen?" })
fetch_content({ urls: ["https://youtube.com/watch?v=abc"], timestamp: "23:41-25:00", frames: 4 })
```
| Parameter | Description |
|-----------|-------------|
| `urls` | Required nonempty array of URL/path strings, including for a single target |
| `prompt` | Question to ask about a YouTube video or local video file |
| `timestamp` | Extract frame(s) — single (`"23:41"`), range (`"23:41-25:00"`), or seconds (`"85"`) |
| `frames` | Number of frames to extract (max 12) |
| `forceClone` | Clone GitHub repos that exceed the 350MB size threshold |
### get_search_content
Retrieve stored content from previous searches or fetches. Content over 30,000 chars is truncated in tool responses but stored in full for retrieval here.
```typescript
get_search_content({ responseId: "abc123", urlIndex: 0 })
get_search_content({ responseId: "abc123", url: "https://..." })
get_search_content({ responseId: "abc123", query: "original query" })
```
## Capabilities
### GitHub repos
GitHub URLs are cloned locally instead of scraped. The agent gets real file contents and a local path to explore with `read` and `bash`. Root URLs return the repo tree + README, `/tree/` paths return directory listings, `/blob/` paths return file contents.
Repos over 350MB get a lightweight API-based view instead of a full clone (override with `forceClone: true`). Commit SHA URLs are handled via the API. Clones are cached for the session and wiped on session change. Private repos require the `gh` CLI.
### YouTube videos
YouTube URLs are processed via Gemini for full video understanding — visual descriptions, transcripts with timestamps, and chapter markers. Pass a `prompt` to ask specific questions about the video. Results include the video thumbnail so the agent gets visual context alongside the transcript.
Fallback: Gemini Web when browser cookies are enabled → Gemini API → Perplexity (text summary only). Handles all URL formats: `/watch?v=`, `youtu.be/`, `/shorts/`, `/live/`, `/embed/`, `/v/`.
### Local video files
Pass a file path (`/`, `./`, `../`, or `file://` prefix) to analyze video content via Gemini. Supports MP4, MOV, WebM, AVI, and other common formats up to 50MB. Pass a `prompt` to ask about specific content. If ffmpeg is installed, a thumbnail frame is included alongside the analysis.
Fallback: Gemini API (Files API upload) → Gemini Web when browser cookies are enabled.
### Video frame extraction
Use `timestamp` and/or `frames` on any YouTube URL or local video file to extract visual frames as images.
```typescript
fetch_content({ urls: ["..."], timestamp: "23:41" }) // single frame
fetch_content({ urls: ["..."], timestamp: "23:41-25:00" }) // range, 6 frames
fetch_content({ urls: ["..."], timestamp: "23:41-25:00", frames: 3 }) // range, custom count
fetch_content({ urls: ["..."], timestamp: "23:41", frames: 5 }) // 5 frames at 5s intervals
fetch_content({ urls: ["..."], frames: 6 }) // sample whole video
```
Requires `ffmpeg` (and `yt-dlp` for YouTube). Timestamps accept `H:MM:SS`, `MM:SS`, or bare seconds.
### PDFs
PDF URLs are extracted as text and saved to `~/Downloads/` as markdown. The agent can then `read` specific sections without loading the full document into context. Text-based extraction only — no OCR.
### Blocked pages
When Readability fails or returns only a cookie notice, the extension retries via Jina Reader (handles JS rendering server-side, no API key needed), then Gemini URL Context API, then Gemini Web extraction when browser cookies are enabled. Handles SPAs, JS-heavy pages, and anti-bot protections transparently. Also parses Next.js RSC flight data when present.
## How It Works
```
web_search(query)
→ Exa (direct API with key, MCP without) → Perplexity → Gemini API → Gemini Web (if browser cookies enabled)
fetch_content({ urls: [url] })
→ Video file? Gemini API (Files API) → Gemini Web (if browser cookies enabled)
→ GitHub URL? Clone repo, return file contents + local path
→ YouTube URL? Gemini Web (if browser cookies enabled) → Gemini API → Perplexity
→ HTTP fetch → PDF? Extract text, save to ~/Downloads/
→ HTML? Readability → RSC parser → Jina Reader → Gemini fallback
→ Text/JSON/Markdown? Return directly
```
## Commands
### /websearch
Open the search curator directly. Runs searches and lets you review, add, select results, and approve a summary before it is sent back to the agent — no LLM round-trip needed.
```
/websearch # empty page, type your own searches
/websearch react hooks, next.js caching # pre-fill with comma-separated queries
```
Results get injected into the conversation when you approve the summary or click "Send selected results without summary". On timeout, the curator auto-submits and falls back to a deterministic summary if no approved draft is present.
### /curator
Toggle or configure the curator workflow at runtime.
```
/curator # toggle on/off
/curator on # enable curator (summary-review)
/curator off # disable curator (raw results only)
/curator summary-review # explicit workflow
```
Persists to `~/.pi/web-search.json` and takes effect on the next `web_search` call. The curator is off by default. When disabled, `web_search` returns raw results without opening the curator window.
### /search
Browse stored search results interactively. Lists all results from the current session with their response IDs for easy retrieval.
### /google-account
Show the active Google account currently authenticated for Gemini Web. Useful when multiple Chromium profiles exist or `chromeProfile` is set in config.
## Activity Monitor
Toggle with **CTRL+SHIFT+W** to see live request/response activity:
```
─── Web Search Activity ────────────────────────────────────
API "typescript best practices" 200 2.1s ✓
GET docs.example.com/article 200 0.8s ✓
GET blog.example.com/post 404 0.3s ✗
────────────────────────────────────────────────────────────
```
## Configuration
All config lives in `~/.pi/web-search.json`. Every field is optional.
```json
{
"exaApiKey": "exa-...",
"perplexityApiKey": "pplx-...",
"geminiApiKey": "AIza...",
"provider": "exa",
"chromeProfile": "Profile 2",
"allowBrowserCookies": false,
"searchModel": "gemini-2.5-flash",
"summaryModel": "anthropic/claude-haiku-4-5",
"workflow": "summary-review",
"curatorTimeoutSeconds": 20,
"githubClone": {
"enabled": true,
"maxRepoSizeMB": 350,
"cloneTimeoutSeconds": 30,
"clonePath": "/tmp/atomic-github-repos"
},
"youtube": {
"enabled": true,
"preferredModel": "gemini-3-flash-preview"
},
"video": {
"enabled": true,
"preferredModel": "gemini-3-flash-preview",
"maxSizeMB": 50
},
"shortcuts": {
"curate": "ctrl+shift+s",
"activity": "ctrl+shift+w"
}
}
```
`EXA_API_KEY`, `GEMINI_API_KEY`, and `PERPLEXITY_API_KEY` env vars take precedence over config file values. `provider` sets the default search provider: `"exa"`, `"perplexity"`, or `"gemini"`. This is also updated automatically when you change the provider in the curator UI. `workflow` sets the curator mode: `"none"` (default, raw results, no curator) or `"summary-review"` (opens curator with auto-generated summary draft). Enable it in this file or toggle at runtime with `/curator`. `chromeProfile` overrides the Chromium profile directory used for Gemini Web cookie lookup. `allowBrowserCookies` enables Chromium cookie extraction for Gemini Web; it defaults to `false` to avoid surprise macOS Keychain prompts. You can also set `PI_ALLOW_BROWSER_COOKIES=1`. `searchModel` overrides the Gemini API model used by `web_search` without changing URL, YouTube, or video extraction defaults. `summaryModel` sets the default model used for generating summary drafts in the curator UI (e.g. `"anthropic/claude-haiku-4-5"` or `"openai-codex/gpt-6.1-sol"`). Only models available in your model registry are eligible; if the configured model is unavailable, the default falls back to the built-in preference list. `curatorTimeoutSeconds` controls the initial curator idle timeout (default `20`, max `600`); users can still adjust the timer in the curator UI.
### Shortcuts
Both shortcuts are configurable via `~/.pi/web-search.json`:
```json
{
"shortcuts": {
"curate": "ctrl+shift+s",
"activity": "ctrl+shift+w"
}
}
```
Values use the same format as pi keybindings (e.g. `ctrl+s`, `ctrl+shift+s`, `alt+r`). Changes take effect on next pi restart.
Set `"enabled": false` under any feature to disable it. Config changes require a Pi restart.
Rate limits: Perplexity is capped at 10 requests/minute (client-side). Content fetches run 3 concurrent with a 30s timeout per URL.
## Limitations
- Chromium cookie extraction for Gemini Web is opt-in via `allowBrowserCookies: true` or `PI_ALLOW_BROWSER_COOKIES=1`. On macOS, enabling it may trigger a Keychain dialog; Linux uses `secret-tool` when available and falls back to Chromium's default password otherwise.
- YouTube private/age-restricted videos may fail on all extraction paths.
- Gemini can process videos up to ~1 hour; longer videos may be truncated.
- PDFs are text-extracted only (no OCR for scanned documents).
- GitHub branch names with slashes may misresolve file paths; the clone still works and the agent can navigate manually.
- Non-code GitHub URLs (issues, PRs, wiki) fall through to normal web extraction.
Files
| File | Purpose |
|------|---------|
| `index.ts` | Extension entry, tool definitions, commands, widget |
| `curator-page.ts` | HTML/CSS/JS generation for the curator UI with markdown rendering |
| `curator-server.ts` | Ephemeral HTTP server with SSE streaming and state machine |
| `summary-review.ts` | Summary prompt construction, model-based draft generation, and deterministic fallback summary |
| `exa.ts` | Exa.ai search provider — direct API and MCP proxy, budget tracking |
| `code-search.ts` | Repository questions via DeepWiki MCP |
| `extract.ts` | URL/file path routing, HTTP extraction, fallback orchestration |
| `gemini-search.ts` | Search routing across Exa, Perplexity, Gemini API, Gemini Web |
| `gemini-url-context.ts` | Gemini URL Context + Web extraction fallbacks |
| `gemini-web.ts` | Gemini Web client (cookie auth, StreamGenerate) |
| `gemini-web-config.ts` | Gemini Web profile and browser-cookie opt-in config |
| `gemini-api.ts` | Gemini REST API client (generateContent) |
| `chrome-cookies.ts` | macOS/Linux Chromium-based cookie extraction (Keychain/secret-tool + SQLite) |
| `youtube-extract.ts` | YouTube detection, three-tier extraction, frame extraction |
| `video-extract.ts` | Local video detection, Files API upload, Gemini analysis |
| `github-extract.ts` | GitHub URL parsing, clone cache, content generation |
| `github-api.ts` | GitHub API fallback for large repos and commit SHAs |
| `perplexity.ts` | Perplexity API client with rate limiting |
| `pdf-extract.ts` | PDF text extraction, saves to markdown |
| `rsc-extract.ts` | RSC flight data parser for Next.js pages |
| `utils.ts` | Shared formatting and error helpers |
| `storage.ts` | Session-aware result storage |
| `activity.ts` | Activity tracking for the observability widget |
| `skills/librarian/` | Bundled skill for library research |