Supercharging Your Daily Workflow
with Aider and Local LLMs


AI Sharing sessions

2025-07-09 @ Taipei City, Taiwan


By Jazz Yao-Tsung Wang
Sr. Director of Engineering, TAO Digital
Chairman, Taiwan Data Engineering Association		 Jazz Wang

Agenda

  • What is Aider?
  • What is Ollama?
  • Why do we need to use local LLM?
  • Install Aider on different platform
    • Google Colab / Cloud Shell
    • Github Codespaces
    • local Windows machine
  • Switch between LLM models
    • Local LLM: Ollama
      • Google Gemma 3 1B
    • Public Free LLM
      • Google Gemini 2.5 Flash
  • How to request Google Gemini API key?
  • Known Use Cases that I use Aider:
    • Reverse Engineering
      • create high-level overview
      • create sequence diagram
    • Meeting Summary
      • MS Teams Live Caption (JSON)
      • MS Stream Transcript file (WebVTT)
    • Monthly Team Achievement Summary
      • from Jira JQL export CSV file
    • rewrite program:
      • from Shell Script to Python script
      • migrate from Selenium to Playwright
    • create unit test

What is Aider?

  • AI Pair Programming in Your Terminal:
    Aider allows you to pair program with Large Language Models (LLMs) directly from your terminal.
  • Works with Existing Codebases:
    Ideal for starting new projects or building upon your current code.
  • Supports Various LLMs:
    Works best with models like Claude 3.7 Sonnet, DeepSeek R1 & Chat V3, OpenAI o1, o3-mini & GPT-4o, and can connect to almost any LLM, including local models.
  • Codebase Mapping:
    Creates a map of your entire codebase to work effectively on larger projects.

https://aider.chat/

Source: https://aider.chat/assets/shell-cmds-small.mp4

Aider Features ( 1 ) - I tried

https://aider.chat/docs/languages.html

hhttps://aider.chat/docs/usage/watch.html
  • Extensive Language Support
    Compatible with over 100 programming languages (Python, JavaScript, Rust, Ruby, Go, C++, PHP, HTML, CSS, etc.).
  • Git Integration
    Automatically commits changes with sensible messages, allowing easy diffing, management, and undoing of AI changes using standard Git tools.
    • Tip: aider --no-auto-commits to disable
    • Tip: /undo to uncommit
  • IDE Integration
    Can be used from within your favorite IDE or editor by adding comments to your code ended by AI! or AI?.
    • Tip: aider --watch-files
  • Web Chat Compatibility
    Can streamline copy/pasting code with LLM web chat interfaces, though it works best with LLM APIs.

Aider Features ( 2 ) - I dit not yet try

  • Multimedia Context
    Supports adding images and web pages to the chat for visual context, screenshots, and reference documentation.
    • (optional) need playwright to fetch web pages
  • Voice-to-Code
    Enables speaking with Aider to request features, test cases, or bug fixes using your voice.
  • Linting & Testing Integration
    Automatically lints and tests code after changes, and can fix problems detected by linters and test suites.

Source: https://aider.chat/docs/usage/lint-test.html

What is Ollama?

  • What is Ollama?
    • open-source tool to run Large Language Models (LLMs) on local machine.
    • Simplifies local LLM deployment and management.
    • Powered by llama.cpp
  • Core Benefits: Local Execution & Control
    • Data Privacy & Data Ownership:
      Keeps sensitive data on your machine.
    • Offline Capability:
      No internet needed after initial download.
    • Reduced Latency:
      Faster interactions without cloud reliance.
    • Cost-Effective:
      Avoids recurring cloud fees.

https://ollama.com/models

Reference: https://www.hostinger.com/tutorials/what-is-ollama

Ollama: Features & Use Cases

  • Key Features
    • Local AI Model Management
      Full control to download, update, delete, and track different versions of models
    • Command-Line (CLI) & GUI Options
      Primarily operates via CLI for precise control
      supports third-party GUIs like Open WebUI for visual interaction.
    • Multi-Platform Support
      Broad compatibility across macOS, Linux, and Windows.
  • Practical Use Cases
    • Building private chatbots and AI applications,e.g. legal, healthcare.
    • Conducting secure, offline AI research.
    • Integrating AI into existing systems with data privacy in mind
      • e.g. ERP, CRM within LAN

Ollama: few useful Tips

  • I pull GGUF format models from LMStudio Community on HuggingFace
    • ollama pull hf.co/lmstudio-community/gemma-3-4B-it-qat-GGUF:Q4_0
  • get context length, embedding length, parameters, quantization
    • ollama show {MODEL}
  • observe System Prompt and Response Template from modelfile
    • ollama show {MODEL} --modelfile

Why do we need to use local LLM?

  • Enhanced Data Privacy & Data Security
    Data stays local, reducing third-party exposure.
  • No Reliance on Cloud Services
    Complete control over infrastructure, greater scalability on local servers.
  • Customization Flexibility
    Tweak models for specific project requirements and tailored datasets.
  • Offline Access
    Work without an internet connection.
  • Cost Savings
    Avoid recurring cloud storage, data transfer, and usage fees.

Guideline: When to use local LLM?

Use Local LLM

When your prompt (input data) contains:

  • Proprietary Business Information
    information created by the organization
    e.g. Company's Source Code, Financial, Patents, etc.
  • Material Nonpublic Information (MNPI)
    organizational information not released to the public
    e.g. internal server name, password, announcement, wiki, etc.
  • Personal Information
    information that can identify an individual.
    e.g. email address, phone, mobile, address, birth, etc.
Use Public LLM
  • translate non-sensitive paragraph
  • summarize non-sensitive paragraph
  • deep research based on web search results
  • brainstorm / plan / generate text with Reasoning LLMs

How to install Aider on different platforms?

Here are PROs and CONs running Aider + Ollama on different environments:

Environment vCPU RAM VRAM Disk PROs CONs Labs
Google Colab 2 12 GB (T4) 15GB 72 GB free T4 GPU time out Open in Colab
GCP Cloud Shell 2 7 GB N/A 4.5 GB free Gemini API Key limited RAM. No GPU. limited Disk. Open in Cloud Shell
Github Codespaces 4 15 GB N/A 16 GB up to 4 cores/16GM RAM/32 GB Disk No GPU Open in GitHub Codespaces
Laptop: Windows 11 ? ? ? ? Offline Development Slow without GPU

Aider: How to switch between local and public LLM models?

  • use /model ollama/{OLLAMA_LIST_MODELS} to use local Ollama model
  • use /model gemini/{TAB key} to use Google Gemini LLMs.
  • use /models {keyword} to list all supported LLM models matched {keyword}.

LLM: How to request Google Gemini API key? 

GEMINI_API_KEY={paste your new key here}

Supercharging Your Daily Workflow with Aider

Here are a list of use cases that I use Aider before:

# Use Case Description Labs
A Reverse Engineering create high-level overview and sequence diagram Open in GitHub Codespaces
B Meeting Summary (1) create meeting summary based on MS Teams Live Caption JSON file Open in GitHub Codespaces
C Meeting Summary (2) create meeting summary based on MS Stream Transcript WebVTT file Open in GitHub Codespaces
D Monthly Team Achievement Summary create Team Achievement Summary based on Jira Export file Open in GitHub Codespaces
E Rewrite Program (1) From Shell Script to Python script Open in GitHub Codespaces
F Rewrite Program (2) Migrate from Selenium to Playwright Open in GitHub Codespaces
G Unit Test Create unit test of existing code Open in GitHub Codespaces

Case A. Reverse Engineering

  • Needs:
    • As a Developer, you always need to read others' source code.
    • As as Developer, you might need to write document for your code.
      • Sequence Diagram / Class Diagram / Package Diagram
  • Requirement:
    • local clone of git repository
  • Example Prompts:
/ask could you give me a high-level overview of this repository?
/ask could you draw a sequence diagram using mermaid.js syntax?
/ask could you draw a class diagram using mermaid.js syntax?
/ask could you draw a package diagram using PlantUML syntax?

Case B. Meeting Summary: From MS Teams Live Caption (JSON file format) 

/ask could you summarize this transcript based on different attendees? highlight action items if possible.

Case C. Meeting Summary: From MS Stream Transcript file (WebVTT file format)

  • Requirement:
    • Create a new Git Repo - git init
    • (Optional) If you don't have permission to download transcript,
      you will need to install Chrome extension (compatible with Edge):
      Transcript Extractor for Microsoft Stream
  • Input:
    • WebVTT transcript file (.vtt)
    • WebVTT-to-JSON transcript file (.json)
  • Known Limitation:
    • context window size (context length) of LLM
  • PS: can use the same on YouTube transcripts
  • Example Prompts:
/ask could you summarize this transcript based on different attendees? highlight action items if possible.
/ask could you separate this transcript based on topics and summarize each topics based on different attendees? please add action items if possible.

based on the transcript, could you create a report for "{MASKED}" project, the following sections are required:
  - Project Overview
  - Business Objective/Problem Statement
  - How will the product/project benefit the dealerships/end customers

Case D. Jira Summary - Monthly Team Achievement Summary

  • Requirement:
    • Create a new Git Repo - git init
  • Input:
    • Jira JQL export CSV file
  • Known Limitation:
    • context window size (context length) of LLM
  • Example Prompts:
/ask could you summarized the following jira exported "summary" into 3 bullet points within 100 words?

Case E. Rewrite Program (1) : From Shell Script to Python script

  • Requirement:
    • local clone of git repository
  • Example Prompts:
/read-only src/download-sftp-files.sh
/run touch src/download-sftp-files.py
/git add src/download-sftp-files.py
/add src/download-sftp-files.py

Case F. Rewrite Program (2) : Migrate from Selenium to Playwright

  • Requirement:
    • local clone of git repository
  • Example Prompts:
/add src/get-pages.py
/ask could you rewrite `get-pages.py` using playwright to replace selenium?

G. Create Unit Tests

  • Requirement:
    • local clone of git repository
  • Example Prompts:
/read-only src/get-pages.py
/ask could you create a unit test program for `get-pages.py`?

Summary

  • Let's recap what we covered today
    • What is Aider
    • What is Ollama
    • Data Privacy & Data Control
    • Local LLMs v.s. Public LLMs
    • Different ways to install Aider on different OS and Platform
    • 8 Use Cases to boost your productivity.
  • Jazz's Suggestion
    • Practice "Prompt Engineering"
    • Try to get familiar with a AI Code Assistant based on your preference!

Q & A

Learn by Doing

Supercharging Your Daily Workflow
with Aider and Local LLMs


Contact Jazz Yao-Tsung Wang
https://www.linkedin.com/in/jazzwang/		 Jazz Wang