6.0 KiB
created, modified, type, tags, aliases
| created | modified | type | tags | aliases | |||
|---|---|---|---|---|---|---|---|
| 2026-05-23 10:07 | 2026-05-23 10:07 | note |
|
AI Tools to try
🌐 Pi-Agent Browser Tools Matrix
| Extension | Command / Tool | Strategy | Context Footprint | Live Auth Session? |
|---|---|---|---|---|
pi-browser-harness |
/browser-setup |
Accessibility Tree (AX) over direct CDP | Minimal | Yes (Uses your active Chrome profile) |
pi-agent-browser-native (@picassio/...) |
agent_browser |
Native wrapper for agent-browser CLI |
Optimized / Compact | Isolated Profile (Saves cookies) |
@amaster.ai/pi-browser-use |
browser_ tools |
Wraps official Chrome DevTools MCP (chrome-devtools-mcp), browser_ prefix |
Moderate | Isolated default; can attach to running Chrome |
@bytesbrains/pi-textbrowser |
text_browser |
Headless DOM + OCR text maps (no image tokens) | Ultra-Low | No |
camofox-browser (via @askjo/camofox-browser-mcp or camofox-web-search-pi) |
camofox_* tools |
Standalone anti-detection browser server (Camoufox/Firefox fork) — REST API :9377 | Low (AX snapshots) | Cookie import + persisted sessions; VNC login |
Verified 2026-09-08:
pi-textbrowser(old name) was unpublished from npm 2026-07-21; same project lives on as@bytesbrains/pi-textbrowser. Matrix updated to reflect real, currently-installable packages.
🚀 Quick Implementation Snippets
Option A: pi-browser-harness (Recommended for Authenticated Web Apps)
Use this if you want Pi to look at sites you are already logged into (Jira, GitHub, Gmail) without passing raw API keys or passwords.
# 1. Install the extension in your environment
pi install npm:pi-browser-harness
# 2. Expose the Chrome DevTools Protocol port (Run in a separate terminal)
# On macOS:
/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --remote-debugging-port=9222
# On Windows:
start chrome.exe --remote-debugging-port=9222
# 3. Inside your Pi TUI, configure the profile connection
/browser-setup
/browser-profile default
Option B: pi-agent-browser-native (Recommended for Clean UI and QA)
Use this if you want an isolated browser session dedicated specifically to searching docs or running end-to-end interface testing.
# 1. Install the core upstream CLI tool globally
npm install -g agent-browser
# 2. Add the optimized native package fork via Pi
pi install npm:@picassio/pi-agent-browser-native
# 3. Reload your environment variables and config
/reload
# 4. Prompt the agent directly to verify installation:
# "Use the agent_browser tool to open https://example.com and take an interactive snapshot."
🦊 Option C: camofox-browser (Stealth Scraping / Bot-Walled Sites)
Use this if you need access to pages that block normal playwright/headless browsers (Cloudflare challenges, bot detection) — e.g. heavy research scraping, YouTube transcripts, price/product research.
- What it is: a standalone REST server (localhost:9377) wrapping Camoufox — a Firefox fork with C++-level fingerprint spoofing. 10.4K stars, MIT, very active (v1.14.0 Aug 2026 / npm 2.4.7), created Jan 2026 (young project).
- Not a pi-native tool by default — it's a server you run (npm/Docker/Fly/Railway), then reach from agents via its MCP server (
@askjo/camofox-browser-mcp) or REST. There is a small Pi extension (camofox-web-search-pi) for the web-search side. - Notable features: accessibility snapshots ~90% smaller than HTML, stable element refs (
e1,e2…), session isolation + persistence (~/.camofox/profiles/), Netscape cookie import for authenticated sites, VNC interactive login, YouTube transcripts (no API key), 15 search macros (@google_search,@youtube_search,@amazon_search,@reddit_subreddit…), screenshots, proxy + GeoIP, structured JSON extraction, Playwright traces, OpenAPI docs. - NixOS: explicit
CAMOUFOX_EXECUTABLE=/nix/store/.../camoufox-binsupport; otherwise postinstall auto-fetches the Camoufox binary. Docker is the cleanest route on NixOS. - Caveats (assessed): young + fast-growing repo (hype factor), ~200MB Camoufox binary, crash telemetry ON by default (anonymized, posts GitHub issues; opt-out
CAMOFOX_CRASH_REPORT_ENABLED=false), 131 open issues, and it's a marketing vehicle for the "jo" AI agent. Overkill unless you actually hit bot walls.
# Option C quickstart (Docker — cleanest on NixOS)
cd camofox-browser && make up # clone https://github.com/jo-inc/camofox-browser
# connect from pi via MCP:
# npx @askjo/camofox-browser-mcp (stdio MCP server proxying :9377)
Understand Anything
Turn any codebase knowledge base into an interactive knowledge graph https://github.com/Lum1104/Understand-Anything
Skybridge MCP app builder
Babysitter
source_url: https://github.com/a5c-ai/babysitter
Enforce obedience to agentic workforces. Manage extremely complex workflows through deterministic, hallucination-free self-orchestration.
flue framework
This lets you make an AI Agent workflow which does not need a human to work with it. Seems good. Might need to re-think my set up.
https://www.youtube.com/watch?v=n5cYS6KuyK8
Scraping python web scrapper
Its good because it fixes itself when classes, divs etc change.
https://www.youtube.com/watch?v=q-uj7wk0LRI
https://github.com/d4vinci/Scrapling
TencentDB Agent Memory
https://github.com/TencentCloud/TencentDB-Agent-Memory
4-tier auto-memory for AI agents. Captures conversations → extracts atomic facts → groups into scenarios → builds persona profile. Solves agent amnesia — the agent remembers your preferences, conventions, and past sessions automatically. 11K stars, MIT. v2 (Docker multi-service) in beta.
⏳ Wait for v2 stable. Evaluate whether conversation-memory benefit justifies Docker overhead given we already have engram + codebase-memory.
HUNK - git differences viewer
https://github.com/modem-dev/hunk https://www.youtube.com/watch?v=-4fJbIF8WAs