Build an AI-Powered CLI Assistant

Build a terminal assistant that answers questions about your local files using an LLM.

Intermediate

3 hours

4 lessons

Course Content

31 chapters across 4 lessons

Index Any File Type

Replace the PDF-only pipeline with one that reads any text-based file from a folder

8 chapters

Progress0%

Clean Up and List Files

Continue

Remove development helpers and add a directory walker that filters by extension

Read a File's Content

Add a function that reads any text file and silently skips files it can't read

Collect Text from All Files

Add the first part of the folder indexer: list files, read each one, and accumulate results

Chunk into Structured Format

Rewrite the indexer to produce chunks with source filenames attached

Extract to files.py

Move the file I/O functions into a dedicated module

Update app.py

Remove the file I/O functions from app.py and rewrite main() to accept a folder

Run the Indexer

Run the updated app.py against a docs folder and verify the output

Lesson Complete

You finished Index Any File Type

Add an Interactive Loop

Replace the one-shot CLI with a REPL that stays alive for multiple questions

8 chapters

Progress0%

One-Shot vs. Interactive

Understand why a persistent loop is better than restarting the program for each question

Build the Loop

Add the chat_loop function with a while True input cycle

Update Callers for Dict Chunks

Fix build_prompt to extract text from dict chunks returned by index_folder

Wire the Pipeline

Connect search, build_prompt, generate_answer, and print inside chat_loop

Tell the Model What It Knows

Inject the list of indexed files into the prompt so the model can answer meta questions

Update main()

Call chat_loop at the end of main() to hand off to the interactive session

Ask Anything

Try the working interactive assistant with sample questions

Lesson Complete

You finished the Interactive Loop

Remember the Conversation

Add conversation history so follow-up questions work correctly

5 chapters

Progress0%

LLMs Are Stateless

Understand why follow-up questions fail and how conversation history fixes them

Format History

Write a function that converts the conversation history list into a readable string for the prompt

Pass History to the Prompt

Update build_prompt to accept conversation history and embed it between the context and the question

Track Turns

Update chat_loop to maintain conversation history and pass it to build_prompt on every turn

Lesson Complete

You finished Conversation History

Streaming & Commands

Replace the blocking response with streaming output, then add slash commands and @filename routing to give users control over the assistant.

10 chapters

Progress0%

Why Streaming Matters

Understand the difference between waiting for a complete response and printing tokens as they arrive

Build the Streaming Function

Replace the blocking generate_answer with stream_answer that prints tokens as they arrive

Switch to Streaming in the Loop

Delete generate_answer and replace its call in chat_loop with stream_answer

Rename messages to history

Rename the messages variable in chat_loop to history to match build_prompt's parameter name

Handle /quit and /files

Add slash command detection to chat_loop with /quit to exit and /files to list indexed files

Handle /new and /help

Add /new to reset conversation history and /help to print available commands

Detect the @filename Syntax

Write resolve_file_reference to extract @filename from a question and return the matching chunks

Use File Routing in the Loop

Update chat_loop to call resolve_file_reference before deciding whether to run vector search

Try the Commands

Run the assistant against the docs folder and test slash commands and @filename routing

What Your Assistant Can Do

Review the complete AI CLI assistant and the interaction patterns it supports