Skip to content

ROADMAP.md

Latest commit

 

History

History
661 lines (547 loc) · 20.5 KB

ROADMAP.md

File metadata and controls

661 lines (547 loc) · 20.5 KB
Raw

ProxyMe - Comprehensive Development Roadmap

Version: 2.0.0 (Planning)
Current Version: 1.0.3
Last Updated: October 20, 2025
Status: 🚀 Active Development

📋 Table of Contents

  1. Project Vision
  2. Current Status
  3. Version 2.0.0 - Major Redesign
  4. Future Versions
  5. Long-Term Vision
  6. Known Issues
  7. Ideas & Explorations

🎯 Project Vision

Mission: Provide a free, open-source alternative to expensive JetBrains AI subscriptions by enabling developers to use their own AI API keys (DeepSeek, Perplexity, etc.) directly in Rider IDE.

Core Values:

  • 🆓 Free & Open Source
  • 🔐 Privacy First (API keys stored locally, encrypted)
  • 🎨 Native Integration (No custom UI, works with Rider AI Assistant)
  • 🚀 Easy Setup (Preset templates, one-click configuration)
  • 🌍 Community Driven (Open to contributions)

✅ Current Status (v1.0.3)

What's Working

✅ Core Functionality

  • Node.js proxy server launching and management
  • Shared dependency architecture (~/.proxyme/proxy/)
  • Health check endpoint with model status
  • OpenAI-compatible API endpoints
  • Streaming support for real-time responses
  • Automatic Node.js path detection

✅ Rider IDE Integration

  • Native AI Assistant Chat mode integration
  • 7 AI models available (DeepSeek + Perplexity)
  • Settings panel with proxy controls
  • Status bar widget
  • Tool window with live logs
  • Launch/Stop/Restart/Health Check buttons

✅ Models Supported

  1. deepseek-chat - General purpose (DeepSeek-V3.1)
  2. deepseek-reasoner - Advanced reasoning with CoT
  3. sonar - Lightweight search with grounding
  4. sonar-pro - Advanced search
  5. sonar-reasoning - Real-time reasoning
  6. sonar-reasoning-pro - Precise reasoning (DeepSeek-R1)
  7. sonar-deep-research - Exhaustive research

✅ UX Features

  • Inline dependency installer with progress
  • One-time shared dependency setup
  • Status bar widget (green/red indicator)
  • Real-time log viewer in tool window
  • Auto-launch on Rider startup (optional)
  • Proxy helper script for debugging

✅ Documentation

  • Comprehensive README.md
  • DEBUGGING_GUIDE.md
  • SUCCESS_AND_NEXT_STEPS.md
  • DEEPSEEK_API_ENDPOINTS.md
  • API_ENDPOINT_SUMMARY.md
  • RIDER_INTEGRATION_MODES.md
  • CHEATSHEET.md
  • Proxy helper script

⚠️ Known Limitations

  • Quick Edit mode has patch format issues (AI returns conversational text instead of XML patches)
  • FIM (Fill-In-Middle) code completion not supported yet
  • Template system exists but not fully functional
  • API keys stored in separate fields (not integrated with model configuration)
  • No preset templates for easy setup
  • Can't assign models to specific Rider AI contexts

🚀 Version 2.0.0 - Major Redesign

Target Release: Q1 2025
Status: Planning Phase
Breaking Changes: Yes (settings format migration required)

🎨 Settings UI Redesign

Move API Keys Into Model Configuration

  • ✨ NEW: API Key column in Model Configuration table
  • ✨ NEW: Each model has its own API key (encrypted)
  • 🗑️ REMOVE: Separate API key fields at top of settings

Template System Overhaul

  • ✨ NEW: Working Load/Save template buttons
  • ✨ NEW: Template storage in ~/.proxyme/templates/
  • ✨ NEW: 7 preset templates packaged with plugin:
    1. Recommended (DeepSeek Chat + Sonar) ⭐
    2. All DeepSeek Models
    3. All Perplexity Models
    4. DeepSeek Only
    5. Research Setup (Reasoner + Deep Research)
    6. Fast Setup (optimized for speed)
    7. Complete (all 7 models)
  • ✨ NEW: Template preview before loading
  • ✨ NEW: Include/exclude API keys when saving templates
  • ✨ NEW: Share templates with team (export/import)

Rider AI Context Assignment

  • ✨ NEW: Assign models to specific Rider AI contexts:
    • Chat - General conversations in AI Assistant
    • Inline Edit - Quick Edit / Modify Selected Code
    • Auto Apply - Automatic suggestions
    • Commits - Git commit message generation
    • Naming - Variable/function naming suggestions
  • ✨ NEW: Multi-select checkboxes for context assignment
  • ✨ NEW: Default assignments based on model capabilities
  • ✨ NEW: Validation (at least one model per context)

Enhanced Model Configuration Table

  • ✨ NEW: Columns:
    • Enabled (checkbox)
    • Model Name
    • Provider (DeepSeek/Perplexity)
    • API Key (encrypted, shown as sk-••••••)
    • Assigned Contexts (Chat, Edit, +2)
    • Category (Core/Research/Fast)
  • ✨ NEW: Inline editing for API keys
  • ✨ NEW: Context assignment dialog
  • ✨ NEW: Drag-and-drop row reordering
  • ✨ NEW: "Load Preset..." dropdown for quick setup

🔐 Security Enhancements

  • ✨ NEW: API key encryption using Rider's PasswordSafe
  • ✨ NEW: Never log API keys (show as masked)
  • ✨ NEW: Encrypted template storage (optional)
  • ✨ NEW: Secure env file generation (chmod 600)

📦 Template System Details

Template File Format (JSON)

{
  "name": "Recommended Setup",
  "description": "DeepSeek Chat + Sonar",
  "version": "2.0",
  "isPreset": true,
  "models": [
    {
      "modelName": "deepseek-chat",
      "apiProvider": "deepseek",
      "apiKey": "",
      "enabled": true,
      "assignedContexts": ["chat", "edit", "commits", "naming"]
    }
  ],
  "proxySettings": {
    "port": 3000,
    "autoLaunch": true
  }
}

Template Storage Structure

~/.proxyme/
├── proxy/                    # Shared proxy files
├── logs/                     # Log files
└── templates/                # NEW
    ├── presets/              # Built-in (read-only)
    │   ├── recommended.json
    │   ├── all-deepseek.json
    │   ├── all-perplexity.json
    │   ├── deepseek-only.json
    │   ├── research-setup.json
    │   ├── fast-setup.json
    │   └── complete.json
    └── user/                 # User-saved
        ├── my-setup.json
        └── work-config.json

🔧 Implementation Phases

Phase 1: Data Model Refactoring (1-2 days)

  • Update ProxyMeSettings.ModelConfig:
    • Add apiKey field (encrypted String)
    • Add assignedContexts field (List)
    • Remove apiKeyName field
  • Remove top-level API key fields from ProxyMeSettings
  • Create RiderAIContext enum
  • Create TemplateService class
  • Implement APIKeyEncryptionService
  • Update ProxyMeEnvFileService to read keys from models

Phase 2: Template System (2-3 days)

  • Create template directory structure
  • Implement TemplateService methods:
    • List<Template> loadPresetTemplates()
    • List<Template> loadUserTemplates()
    • void saveTemplate(Template, String name)
    • Template loadTemplate(String name)
    • void deleteTemplate(String name)
  • Create 7 preset template JSON files
  • Package presets in plugin resources
  • Copy presets to ~/.proxyme/templates/ on first launch

Phase 3: UI Redesign (3-4 days)

  • Remove API key fields from top of settings panel
  • Add API Key column to model table
  • Add Contexts column to model table
  • Create ModelEditDialog with:
    • Model name, provider, endpoint fields
    • API key field (masked input)
    • Context assignment checkboxes
  • Create TemplateLoadDialog:
    • Show preset templates section
    • Show user templates section
    • Template preview panel
    • Search/filter functionality
  • Create TemplateSaveDialog:
    • Name and description fields
    • "Include API keys" checkbox
    • Save location selection
  • Wire up "Load Template", "Save Template", "Delete Template" buttons
  • Add "Load Preset..." dropdown button

Phase 4: Context Assignment (2-3 days)

  • Research Rider AI context configuration API
  • Implement model-to-context mapping
  • Test context switching in Rider AI Assistant
  • Add context validation logic
  • Document which contexts work with ProxyMe

Phase 5: Migration & Testing (2-3 days)

  • Implement automatic settings migration from v1.x
  • Test all 7 preset templates
  • Test template save/load
  • Test API key encryption/decryption
  • Test "Modify Selected Code" with context assignments
  • Update all documentation
  • Create migration guide

Phase 6: Documentation (1-2 days)

  • Create TEMPLATE_GUIDE.md
  • Create MIGRATION_GUIDE_V2.md
  • Create CONTEXT_ASSIGNMENT.md
  • Update README.md with new screenshots
  • Update SUCCESS_AND_NEXT_STEPS.md
  • Create video tutorial for template system

📊 Success Metrics (v2.0.0)

  • Settings migration works for 100% of v1.x users
  • All 7 preset templates load successfully
  • Template save/load roundtrip preserves configuration
  • API keys encrypted and never logged
  • Models correctly assigned to Rider AI contexts
  • "Modify Selected Code" works with assigned models
  • User documentation covers all new features

🎯 Timeline

Estimated Duration: 10-15 days
Target Completion: December 2025

🔮 Future Versions

Version 2.1.0 - Quick Edit Fix

Focus: Solve the Quick Edit patch format issue

Planned Features

  • Response Post-Processing
    • Detect Quick Edit requests (check for patch instructions in system prompt)
    • Parse AI conversational responses
    • Extract code changes
    • Convert to XML patch format: <llm-patch path="...">...</llm-patch>
    • Return formatted patch to Rider
  • Enhanced System Prompts
    • Add stronger XML format examples
    • Include sample patches in prompt
    • Test with DeepSeek-Reasoner (better instruction following)
  • Model-Specific Handling
    • Identify which models naturally follow patch format
    • Route Quick Edit to compatible models
    • Fallback to Chat mode if needed
  • Documentation
    • Document which models work best for Quick Edit
    • Provide workarounds for incompatible models

Success Metrics

  • Quick Edit works with at least 3 models
  • Patch format conversion success rate > 80%
  • Clear user guidance on model selection

Version 2.2.0 - FIM Support (Code Completion)

Focus: Add Fill-In-Middle code completion support

Status: 🔜 Future (Low priority, user-requested feature)

What Is FIM?

  • FIM = Fill-In-Middle code completion
  • Inline autocomplete as you type (like GitHub Copilot)
  • Uses DeepSeek /beta/completions endpoint
  • Different from Chat mode (not conversational)

Planned Features

  • FIM Endpoint Integration
    • Add /beta/completions endpoint support to proxy
    • Implement FIM request/response format
    • Handle prefix + suffix completion
  • Rider IDE Integration
    • Research Rider's inline completion API
    • Determine if third-party completion providers supported
    • May require custom plugin architecture
  • Model Support
    • deepseek-chat (FIM-enabled)
    • Optional: Other FIM-compatible models
  • Settings Configuration
    • Enable/disable FIM feature
    • Assign model for FIM completions
    • Configure max tokens (4K limit)
  • Alternative: Inception Labs Mercury Editor
    • Investigate Mercury Editor integration
    • Compare with DeepSeek FIM
    • Evaluate as alternative option

Implementation Challenges

  • Unknown: Does Rider support third-party inline completion?
  • Complex: Different architecture from Chat mode
  • Alternative: May need custom completion UI
  • Priority: Low (Chat mode is more valuable)

Decision Points

  • Research Rider completion plugin API
  • Evaluate user demand for FIM
  • Determine if separate plugin needed
  • Consider focusing on Chat/Quick Edit instead

Why Not Prioritized?

  1. ✅ Chat mode already works perfectly
  2. ✅ Native Rider integration is our strength
  3. ⚠️ FIM requires complex IDE integration
  4. ⚠️ Rider may not support third-party completion
  5. ⚠️ Limited to DeepSeek models only
  6. 💡 Chat mode covers most use cases

Version 2.3.0 - Advanced Features

Model Management

  • Custom model addition (any OpenAI-compatible API)
  • Model testing (validate API key and endpoint)
  • Model usage statistics (requests, tokens, costs)
  • Model performance metrics (latency, error rates)

Template Enhancements

  • Template marketplace (share with community)
  • Template versioning
  • Template diff (compare configurations)
  • Automatic template updates

Proxy Features

  • Multiple proxy instances (different ports)
  • Proxy performance dashboard
  • Request/response logging (optional, secure)
  • Request rate limiting
  • Cost tracking per model

Version 3.0.0 - Multi-IDE Support

Focus: Expand beyond Rider to other IDEs

Target IDEs

  • IntelliJ IDEA (same plugin base as Rider)
  • VS Code (separate extension)
  • Zed IDE (extension system)
  • Theia IDE (extension system)
  • WebStorm, PyCharm, etc. (JetBrains family)

Architecture

  • Shared proxy (same Node.js proxy for all IDEs)
  • IDE-specific plugins (different integration per IDE)
  • Common configuration (settings sync across IDEs)

🎯 Long-Term Vision: JetBrains Marketplace

Mission Statement

Make AI coding assistance affordable and accessible for all developers by creating a free, open-source alternative to expensive proprietary AI subscriptions.

The Problem Today

  • JetBrains AI costs $10/month
  • OpenAI API subscriptions are expensive
  • Vendor lock-in with proprietary models
  • No control over API keys or privacy

The Solution Tomorrow

  • ✅ ProxyMe Plugin on JetBrains Marketplace
  • ✅ Use your own API keys (DeepSeek, Perplexity, OpenAI)
  • ✅ Pay only for what you use
  • ✅ Full control over data and privacy
  • ✅ Support for multiple AI providers
  • ✅ Native Rider integration (no custom UI)

JetBrains Marketplace Roadmap

Phase 1: Plugin Preparation (v2.0-2.2)

  • Complete settings redesign
  • Implement template system
  • Fix Quick Edit mode
  • Comprehensive documentation
  • Video tutorials

Phase 2: JetBrains Submission (v2.5)

  • Create plugin listing
  • Prepare marketing materials
  • Screenshots and videos
  • Submit to JetBrains Marketplace
  • Address review feedback

Phase 3: Launch & Growth (v3.0+)

  • Public launch announcement
  • Community feedback collection
  • Regular updates and bug fixes
  • Add more AI providers
  • Expand to other JetBrains IDEs

Target Goals (First Year)

  • 🎯 1,000+ downloads
  • 🎯 100+ active users
  • 🎯 4.5+ star rating
  • 🎯 Active community contributions
  • 🎯 Support for 10+ AI models

🐛 Known Issues

Critical (Must Fix for v2.0)

  • Quick Edit Mode Patch Format
    • AI returns conversational text instead of XML patches
    • Rider shows "The patch is incomplete" error
    • Solution: Implement response post-processing (v2.1)
  • Template Load/Save Buttons Don't Work
    • Currently no implementation
    • Solution: Implement TemplateService (v2.0)

High Priority

  • API Keys Not Integrated with Models
    • Separate fields, disconnected UX
    • Solution: Move keys into model configuration (v2.0)
  • No Preset Templates
    • Users must manually configure everything
    • Solution: Package 7 preset templates (v2.0)
  • Context Assignment Missing
    • Can't assign models to specific Rider features
    • Solution: Implement context assignment (v2.0)

Medium Priority

  • Settings migration from v1.x to v2.0
  • API key encryption implementation
  • Template sharing/export functionality
  • Model performance metrics
  • Usage cost tracking

Low Priority

  • FIM (Fill-In-Middle) code completion
  • Multiple proxy instances
  • Dark/light theme icons
  • Plugin keyboard shortcuts
  • Advanced debugging tools

💡 Ideas & Explorations

Security Enhancements

  • Encrypted API Key Storage (High Priority for v2.0)
    • Use Rider's PasswordSafe
    • Or implement dotenvx-style encryption
    • Never commit .env to git
    • Consider HashiCorp Vault integration
    • AWS Secrets Manager support
    • SOPS + git-crypt alternative

Advanced Features

  • Perplexity MCP Support
    • Advanced search tools
    • Similar to Cline integration
  • Headless Plugin Architecture
    • Run ProxyMe as background service
    • IDE-independent proxy
    • Multiple IDE connections
  • Usage Monitoring
    • Fetch real usage from cloud providers
    • Don't self-log (unreliable metrics)
    • DeepSeek usage API
    • Perplexity usage API
  • Distributed Proxy
    • Investigate libP2P framework
    • LocalAI.io for distributed web
    • (Currently LocalAI doesn't support cloud API proxying)

UX Improvements

  • In-IDE troubleshooting panel
    • Mirror DEBUGGING_GUIDE.md in native UI
    • Interactive diagnostics
    • One-click fixes
  • Onboarding wizard
    • First-time setup guide
    • API key acquisition help
    • Model recommendations
  • Model recommendations
    • Suggest best model for task
    • Cost vs performance analysis
    • Real-time model status

Community Features

  • Template marketplace
    • Share configurations
    • Community ratings
    • Template search
  • Plugin contribution guide
    • Clear coding standards
    • Testing requirements
    • PR templates
  • Sponsor/support links
    • GitHub Sponsors
    • Ko-fi integration
    • Community funding

📅 Version History

v1.0.3 (Current) - October 20, 2025

  • ✅ Shared dependency architecture
  • ✅ Inline dependency installer
  • ✅ Status bar widget
  • ✅ Tool window with logs
  • ✅ Node.js path detection
  • ✅ 7 AI models supported
  • ✅ Comprehensive documentation
  • ✅ API endpoint investigation completed

v1.0.2 - October 19, 2025

  • ✅ Fixed npm install PATH issues
  • ✅ Improved dependency installation UX
  • ✅ Added startup notifications

v1.0.1 - October 19, 2025

  • ✅ Basic plugin functionality
  • ✅ Manual dependency installation
  • ✅ Settings panel
  • ✅ Proxy launch controls

v1.0.0 - October 17, 2025

  • ✅ Initial release
  • ✅ Node.js proxy server
  • ✅ DeepSeek and Perplexity support
  • ✅ OpenAI-compatible API

🤝 Contributing

We welcome contributions! Areas where help is needed:

High Priority

  • 🎨 UI/UX design for settings panel
  • 🔐 Security review (API key encryption)
  • 📝 Documentation improvements
  • 🧪 Testing across different environments

Medium Priority

  • 🐛 Bug fixes and issue resolution
  • 🎨 Icon design (dark/light themes)
  • 📹 Video tutorial creation
  • 🌍 Internationalization (i18n)

Exploration

  • 🔬 Rider completion plugin API research
  • 🔬 FIM integration feasibility
  • 🔬 Alternative AI provider support
  • 🔬 Distributed proxy architecture

📞 Questions & Decisions Needed

Immediate (v2.0)

  • Approve settings redesign plan?
  • Priority order for implementation phases?
  • API key encryption method (PasswordSafe vs custom)?
  • Template JSON format finalized?

Short Term (v2.1-2.2)

  • Implement Quick Edit fix in v2.1 or defer?
  • Add FIM support in v2.2 or skip?
  • Focus on Rider only or expand to IntelliJ?

Long Term (v3.0+)

  • Multi-IDE support timeline?
  • JetBrains Marketplace submission timeline?
  • Community template marketplace?
  • Paid features or fully free?

🎯 Next Immediate Steps

  1. ✅ Review Settings Redesign Plan (SETTINGS_REDESIGN_PLAN.md)
  2. ✅ Merge ROADMAP files (This file!)
  3. ⏳ Get approval to begin v2.0 development
  4. ⏳ Start Phase 1: Data model refactoring
  5. ⏳ Create 7 preset template JSON files
  6. ⏳ Begin UI redesign implementation

🎊 Summary

ProxyMe is evolving from a functional prototype (v1.0.3) into a polished, user-friendly plugin (v2.0.0) with:

  • ✨ Redesigned settings UI
  • ✨ Working template system with 7 presets
  • ✨ API keys integrated with model configuration
  • ✨ Model assignment to Rider AI contexts
  • ✨ Enhanced security (encrypted keys)
  • ✨ Better UX (one-click setup with presets)

Long-term vision: A free, open-source alternative to expensive AI coding assistants, available on JetBrains Marketplace, supporting multiple IDEs and AI providers.

Status: 🚀 Ready to begin v2.0.0 development
Timeline: 10-15 days for v2.0.0 completion
Next Milestone: Complete settings redesign and template system

Let's make AI coding assistance accessible for everyone! 🎉