Agentic AI Monorepo

This repository contains multiple, focused examples and mini-projects for building Agentic AI apps with Chainlit and the OpenAI Agents SDK. Each subproject is self-contained and runnable with uv.

Prerequisites

• Python 3.13+
• uv (Rust-based Python package manager) — install from https://astral.sh/uv
• Optional: Playwright browsers for the computer-use demo

Environment variables (.env)

Create a .env file at the repo root or in each project folder that needs it:

GEMINI_API_KEY=your_gemini_key_here
TAVILY_API_KEY=your_tavily_key_here

Notes:

• Gemini models are accessed via the OpenAI-compatible Gemini endpoint in these projects.
• Tavily is used for web search in relevant demos.

How to run projects

Use uv run so dependencies from each project's pyproject.toml and uv.lock are resolved automatically. Always run commands from the project folder unless stated otherwise.

1) Web Search Agent (Chainlit)

• Folder: _Projects/Web_Search_Agent
• Features: tool calling (Tavily web search), context-aware get_info tool, streaming responses

Commands:

cd _Projects/Web_Search_Agent
uv run chainlit run main.py --watch

2) Multi-Agent Code Assistant (Chainlit)

• Folder: _Projects/Code_Assistant_agent
• Features: triage + expert agents (Web, Mobile, Agentic AI), agents-as-tools, Tavily search tool

Commands:

cd _Projects/Code_Assistant_agent
uv run chainlit run main.py --watch

3) Function Calling (Chainlit)

• Folder: 07_Function_Calling
• Features: simple function tools with streaming output

Commands:

cd 07_Function_Calling
uv run chainlit run main.py --watch

4) Model Context (Console)

• Folder: Model_Context
• Features: RunContextWrapper[T] pattern for passing local context to tools

Commands:

cd Model_Context
uv run python main.py

5) Streaming Demo (Console)

• Folder: Model_configuration/05_Streaming
• Features: token streaming via ResponseTextDeltaEvent

Commands:

cd Model_configuration/05_Streaming
uv run python main.py

6) Tavily Search Tool (Chainlit)

• Folder: 09_tools/web_search_tool_in_tavily
• Features: direct Tavily integration and formatted result rendering

Commands:

cd 09_tools/web_search_tool_in_tavily
uv run chainlit run main.py --watch

7) Computer Use Tool (Console + local browser)

• Folder: 09_tools/computer_tools
• Features: local Playwright-driven browser automation via ComputerTool

First-time setup:

cd 09_tools/computer_tools
uv add playwright
python -m playwright install

Run:

uv run python main.py

8) Chainlit Starter (Chainlit)

• Folder: 02_chailit

Commands:

cd 02_chailit
uv run chainlit run app.py --watch

9) MCP Notes (Console)

• Folder: 08_MCP

Commands:

cd 08_MCP
uv run python main.py

Troubleshooting

• Cannot instantiate typing.Union: Ensure tool signatures use RunContextWrapper[T] and access data via Wrapper.context (not Wrapper.data). Avoid conflicting names like from multiprocessing import context.
• Tools not called: Verify Agent(..., tools=[...]) is a list (no trailing commas creating tuples). If using Chainlit, restart the server after code changes.
• Tavily errors: Check TAVILY_API_KEY and network connectivity.
• Gemini model errors: Confirm GEMINI_API_KEY and model name (e.g., gemini-2.5-flash).

Repo notes

• Each folder contains its own pyproject.toml and uv.lock. Running with uv run inside that folder will install and use the correct dependencies.
• Chainlit apps can be accessed in the browser at the URL printed in the terminal after startup.

Working of LLM

Two things are sent to LLM . One thing is System Instruction and another is Tool Call

License and Contact

• License: MIT
• Author: Abdullah Malik
• GitHub: @AbdullahMalik17