applyai_tests

testing ground for all my theories on applied ai

templates - create a template that can be reused for ongoing projects

workflow DOCETL - https://ucbepic.github.io/docetl/best-practices/ when desiding what to create makesure you choose the target datastrucutre first e.g. directory -> doctetl <- data

Table of content

Tests

• clever-deductions: set up tool use [/tooluse]
• composio : composio.io seems to be providing auth.

Infrastructure choices

• all the cli agents will become genral intellegence - never use agents that are not developed by fortune 500 companies os you are building on the ogs - https://deepwiki.com/aws/amazon-q-developer-cli-autocomplete/1-overview - codex - claude-code

• notes on operation

UX choice - I would like to be able to take gui projects and host them on the same domain url/(route:repo) for example - combining (rowboat)[https://github.com/rowboatlabs/rowboat] and (docetl)[https://github.com/ucbepic/docetl] with a shared filesystem and interopterbility using a orchestrating agent - presents a paradigm of agent controlled user defined workflows

• could this be done automatically, what architectural and infra decsions should i make?
  • we could use screen use agents : start vm, download pakages, playwrite automation or nlweb or etc,

Docs

• notes on documentation
• set up living documentaion https://platejs.org/docs

keywords - tool use clever deductions

task decomposition

computer use- https://www.hud.so/ https://www.trycua.com/ https://developers.cloudflare.com/changelog/2025-05-28-playwright-mcp/

https://www.hcompany.ai/surfer-h

Pages

Clever Deductions

create a mcp chat interface the interface should include the functionality of;

• A "Hub" - UX of urls - a user can oneclick install a mcp server from a list, install there own tool to a mcp server, and recieve user.tooluse.com/{user:tool_key}/sse