Skip to content

secypher/vaki

Repository files navigation

VÄKI - Your AI Coding Assistant

Stop manually implementing GitHub issues. Let AI do it for you.

VÄKI reads your GitHub issues, implements the changes, and creates pull requests automatically. Think of it as having a junior developer who works 24/7 and never gets tired.


What Does It Do?

The Simple Version:

  1. You have a GitHub issue that needs implementation
  2. Run one command: python main.py run myproject 42
  3. VÄKI implements the changes and creates a PR
  4. You review and merge

Real Example:

Issue #42: "Add dark mode toggle to settings"

$ python main.py run myproject 42 --mode=openai

... 3 minutes later ...

✅ Pull Request created: https://github.com/org/repo/pull/123
   - 3 files changed
   - All tests passing
   - Ready for review

Why Use VÄKI?

The Problem

  • You have a backlog of small bugs and features
  • You spend hours on repetitive implementations
  • Context switching between tickets is exhausting
  • Junior developers need constant code review

The Solution

  • VÄKI handles routine implementations automatically
  • You focus on architecture and complex features
  • Consistent code quality (follows your standards)
  • Safe rollback if something goes wrong

Real Impact:

  • Teams save 5-10 hours per week on routine tasks
  • Faster time to resolution for simple bugs
  • More time for creative problem-solving

Quick Start (5 minutes)

1. Install

# Clone and setup
git clone https://github.com/your-org/vaki
cd vaki
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Configure

Create .env file:

GITHUB_TOKEN=ghp_your_token_here
OPENAI_API_KEY=sk_your_key_here

Create projects/myproject.yml:

name: myproject
github:
  repo: your-org/your-repo
  base_branch: main
filters:
  assignee: your-username
  state: open
context: contexts/myproject.md  # Optional

3. Run

# Process a specific issue
python main.py run myproject 42 --mode=openai

# Or process all assigned issues
python main.py run myproject --mode=openai

That's it! VÄKI will implement the issue and create a PR.

Want more details? See QUICK_START.md for a complete guide.


How It Works

VÄKI has three modes depending on how much control you want:

🤖 OpenAI Mode (Recommended)

Fully automated - Just run and review the PR later

python main.py run myproject 42 --mode=openai

What happens:

  1. Reads your GitHub issue
  2. Analyzes your codebase to understand the project
  3. Generates multiple implementation strategies
  4. Picks the best one and implements it
  5. Runs tests and quality checks
  6. Creates a pull request

Best for: Bug fixes, small features, routine tasks

Time saved: 30 min - 2 hours per issue


👤 Claude Mode (Human-in-the-Loop)

Semi-automated - You guide, Claude helps

python main.py run myproject 42

What happens:

  1. Fetches issue and creates branch
  2. Opens Claude Code CLI with full context
  3. You work with Claude to implement
  4. Creates PR when you're done

Best for: Complex features, architectural changes, learning


📋 Manual Ticket Mode (Multi-Tool Teams)

Process tickets from anywhere - Slack, email, Jira, etc.

# Copy message from Slack
python main.py ticket myproject --source=slack
# Paste message, press Ctrl+D
# PR created!

What happens:

  1. Paste text from any source (Slack, email, Jira)
  2. VÄKI parses it like a GitHub issue
  3. Implements and creates PR

Best for: Ad-hoc requests, teams using multiple tools

Learn more: MANUAL_TICKETS.md


Configuration

Basic (Good Enough)

This is all you need to get started:

# projects/myproject.yml
name: myproject
github:
  repo: your-org/your-repo
  base_branch: main
filters:
  assignee: your-username
  state: open

Advanced (Production-Ready)

Add these for enterprise features:

# Quality control
quality:
  mode: "standard"  # Ensures tests pass, code is clean

# Budget limits
resources:
  daily_cost_limit: 50.00      # Max spend per day
  per_issue_cost_limit: 10.00  # Max spend per issue

# Smart features
ticket_analysis:
  enabled: true  # Checks if issue is clear enough
implementation:
  multi_strategy: true  # Tries multiple approaches
learning:
  enabled: true  # Gets better over time

See example: projects/example-enhanced.yml


Common Questions

Is it safe?

Yes! VÄKI:

  • ✅ Never pushes to main directly (always creates PRs)
  • ✅ Runs your tests before creating PRs
  • ✅ Can rollback changes if something goes wrong
  • ✅ You review everything before merging

How much does it cost?

Typical costs with GPT-5.0:

  • Small bug fix: $1-2
  • Medium feature: $3-6
  • Large feature: $8-15

You can set budget limits to control costs.

Will it replace developers?

No! VÄKI is best at:

  • ✅ Routine bug fixes
  • ✅ Simple feature implementations
  • ✅ Following existing patterns

You're still needed for:

  • ❌ Architecture decisions
  • ❌ Complex algorithms
  • ❌ Creative solutions
  • ❌ Code review and quality

Think of it as an assistant, not a replacement.

What if it makes a mistake?

  1. You review the PR before merging (always!)
  2. If wrong, just close the PR
  3. VÄKI learns from feedback
  4. Budget limits prevent runaway costs

Can I use my own coding standards?

Yes! Create contexts/myproject.md with your standards:

# Coding Standards
- Use TypeScript strict mode
- Write tests for all features
- Follow our component patterns
- Use Tailwind for styling

VÄKI will follow your rules.


Features

Quality & Safety

  • Quality Gates - Won't create PRs with critical issues
  • Checkpoints - Can rollback if implementation fails
  • Incremental Validation - Catches errors early
  • Test Execution - Runs your test suite

Smart Implementation

  • Ticket Analysis - Checks if requirements are clear
  • Multi-Strategy - Tries different approaches
  • Codebase Understanding - Analyzes your project structure
  • Learning System - Improves from past implementations

Cost Control

  • Budget Limits - Daily and per-issue caps
  • Cost Estimation - Know the cost before running
  • Usage Tracking - See exactly what you're spending

Debugging

  • Comprehensive Logs - Everything is recorded
  • Debug Bundles - Complete info for failed runs
  • Metrics Tracking - Success rates, costs, time

Documentation

Get Started:

Learn More:

For Developers:


Requirements

  • Python 3.8+
  • Git
  • GitHub account with API token
  • OpenAI API key (for OpenAI mode)
  • Claude Code CLI (for Claude mode, optional)

Examples

Example 1: Fix a Bug

# Issue #42: "Login button is misaligned on mobile"
python main.py run myproject 42 --mode=openai

# 2 minutes later...
# ✅ PR created with CSS fix

Example 2: Add a Feature

# Issue #55: "Add export to CSV button"
python main.py run myproject 55 --mode=openai

# 5 minutes later...
# ✅ PR created with new export feature

Example 3: Slack Request

# Someone asks in Slack: "Can we add dark mode?"
python main.py ticket myproject --source=slack
# Paste the message, press Ctrl+D
# ✅ PR created!

Example 4: Batch Processing

# Process all assigned issues
python main.py run myproject --mode=openai

# Sit back and watch VÄKI work through your backlog

What's New in v2.0

Major Improvements:

  • 🎯 50-70% fewer failed implementations (ticket analysis)
  • 🛡️ 100% enforcement of quality standards (quality gates)
  • 💰 Predictable costs (budget tracking)
  • 🔄 Safe rollback (checkpoints)
  • 📊 Continuous learning (gets better over time)

New Features:

  • Multi-strategy implementation
  • Real-time validation
  • Cost estimation and tracking
  • Codebase understanding
  • Comprehensive logging

Fully backward compatible - existing setups work unchanged.


Support & Contributing

Questions? Open an issue on GitHub Found a bug? Create an issue with logs Want to contribute? PRs welcome!


License

[Your License Here]


Credits

Built with ❤️ for developers who have better things to do than implement routine tasks.

Powered by:

  • OpenAI GPT-5.0 (for automated implementation)
  • Claude (for human-guided implementation)
  • Your coding standards and best practices

Ready to save hours each week? Start with QUICK_START.md 🚀

About

Väki AI is a personal agentic AI solution to help software developers complete project work faster

Resources

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors