Files
obsidian-vault/300 areas/350 AI/AI Tools to try.md

105 lines
3.5 KiB
Markdown

---
created: 2026-05-23 10:07
modified: 2026-05-23 10:07
type: note
tags:
- ai
- tool
- tools
aliases: []
---
# [[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`** | `agent_browser` | Native wrapper for `agent-browser` | Optimized / Compact | Isolated Profile (Saves cookies) |
| **`@amaster.ai/pi-browser-use`** | `browser_` tools | Multi-tab script automation (`browser-use`) | Moderate | Isolated Session |
| **`pi-textbrowser`** | `text_browser` | DOM to Compressed Markdown / Text | **Ultra-Low** | No |
---
### 🚀 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.
```bash
# 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.
```bash
# 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."
```
### 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://flueframework.com/
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