What's New

NanoGPT Local Mode Authentication

• Fixed 402 "Insufficient balance" errors in local mode by properly passing NANOGPT_API_KEY into Docker container
• Populate authVars from environment variables when running in local mode (GOENV=development and LOCAL_MODE=1)
• Enables seamless NanoGPT subscription usage in self-hosted deployments

XML Streaming Improvements

• Hardened stream processor to prevent incomplete <PlandexFinish tags from appearing in output
• Added stop sequence variants (<PlandexFinish/>, <PlandexFinish />, <PlandexFinish>) for better model compatibility
• Implemented split-chunk detection to handle finish tags that arrive across multiple streaming chunks
• Only append sanitized content to persisted reply to ensure clean final output

Commit Message Generation

• Tightened XML system prompt for commit message generation to enforce exact <commitMsg> element format
• Reduces "No commitMsg tag found in XML response" errors when using XML-based models

Model Packs

• Added nanogpt-subscription-only model pack featuring GLM-4.6 and DeepSeek R1 models
• Optimized for users with NanoGPT subscription plans

Technical Changes

Modified Files:

• app/docker-compose.yml - Pass NANOGPT_API_KEY environment variable to server container
• app/server/handlers/client_helper.go - Populate authVars from environment in local mode
• app/server/model/plan/tell_stream_processor.go - Harden XML tag sanitization and split-chunk handling
• app/server/model/plan/tell_exec.go - Add stop sequence variants, remove debug logging
• app/server/model/prompts/describe.go - Tighten commit message XML prompt
• model-packs/nanogpt-subscription-only.json - New model pack for NanoGPT subscription users

Upgrade Notes

For local/self-hosted deployments:

1. Ensure NANOGPT_API_KEY is set in your environment before starting Docker services
2. Rebuild containers: cd app && docker compose up --build -d

No breaking changes. Existing configurations will continue to work.

Full Changelog: v3.0.0...v3.0.1

🔍 Debug Logging Enhancement

✨ New Features

Debug Logging for Provider Selection

• Added comprehensive debug logging to trace NanoGPT provider selection
• Logs UseNanoGPTSubscription flag value during provider filtering
• Shows which providers are being skipped/added during selection process
• Helps troubleshoot provider selection and subscription mode issues

🛠️ Technical Details

Files Modified:

• app/shared/ai_models_providers.go: Added debug log statements in GetProvidersForAuthVarsWithModelId()

Debug Output Includes:

• Current value of UseNanoGPTSubscription flag
• Provider being evaluated
• Whether regular NanoGPT is skipped in favor of subscription
• Whether subscription provider is skipped when user prefers balance mode
• Which providers are added to the final list

🎯 Benefits

• Troubleshooting: Easier to diagnose provider selection issues
• Transparency: Clear visibility into which endpoint is being used
• Debugging: Helps identify configuration problems with NanoGPT billing modes

📋 Usage

To see debug logs:

cd ~/plandex-server/app
docker compose logs -f plandex-server | grep DEBUG

🔄 Upgrade

For local installations:

plandex update-server

Or manually:

cd ~/plandex-server
git pull origin main
cd app
docker compose down
docker compose build --no-cache plandex-server
docker compose up -d

Related: Works with CLI v2.2.9 NanoGPT subscription mode support

What's new:\n- Fix: CLI resolves auth before fetching OrgUserConfig (prevents 'Auth not loaded' on /org_user_config)\n- NanoGPT: connect/disconnect works in local mode; subscription mode toggle is reliable\n- Eliminates 402 Insufficient balance when using subscription mode\n\nInstall: see assets below for your platform (Linux/macOS, amd64/arm64).

🚀 Enhanced NanoGPT Setup with Subscription Support

This release improves the installation experience for NanoGPT users by adding support for choosing between subscription and balance billing modes during setup.

✨ New Features

• Interactive billing mode selection during NanoGPT setup
• Clear explanation of subscription mode (fixed monthly cost) vs balance mode (pay-per-use)
• Guidance on switching modes with existing CLI commands
• Enhanced provider description to mention billing options
• Improved user experience for choosing the appropriate billing model

🔧 What's Changed

• The install script now explains NanoGPT's two billing modes during setup
• Users can choose their preferred billing mode during installation
• Added references to plandex connect-nanogpt and plandex disconnect-nanogpt commands
• Enhanced guidance for both new and existing NanoGPT users

💡 For NanoGPT Users

After installation, you can switch between billing modes:

• plandex connect-nanogpt - Enable subscription mode (recommended for heavy usage)
• plandex disconnect-nanogpt - Enable balance mode (pay-per-use)

📚 Documentation

For more details about model providers and configuration, visit: https://docs.plandex.ai/models/model-providers

🚀 NanoGPT Subscription Mode Support

✨ New Features

NanoGPT Billing Mode Selection

• Added support for NanoGPT subscription mode vs balance mode
• New CLI commands:
  • plandex nanogpt - Check current NanoGPT configuration
  • plandex connect-nanogpt - Enable subscription mode
  • plandex disconnect-nanogpt - Switch to balance mode
• Automatic endpoint selection based on user preference:
  • Subscription mode: Uses /api/subscription/v1/ endpoint
  • Balance mode: Uses /api/v1/ endpoint (default)

🐛 Bug Fixes

Resolved Balance Depletion Issue

• Fixed issue where NanoGPT subscription users were charged against balance instead of subscription allowance
• Users can now properly utilize their NanoGPT subscription benefits

🔧 Technical Improvements

Provider Architecture Enhancement

• Added ModelProviderNanoGPTSubscription provider variant
• Enhanced provider filtering logic in GetProvidersForAuthVarsWithModelId
• Added UseNanoGPTSubscription preference in OrgUserConfig
• Maintains backward compatibility with existing configurations

📋 Supported Models

All 13 NanoGPT-supported models now work with both billing modes:

• OpenAI: o3, o4-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano
• Anthropic: claude-sonnet-4-5, claude-3.5-sonnet, claude-3.5-haiku
• Google: gemini-1.5-pro, gemini-2.5-pro, gemini-2.5-flash
• DeepSeek: deepseek-chat, deepseek-r1

🎯 Impact

• Before: NanoGPT subscription users experienced unexpected balance charges
• After: Users can choose their preferred billing mode and use subscription allowance properly

Full Changelog: cli/v2.2.6...cli/v2.2.7

🚀 Enhanced Installation & Upgrade Process

This release significantly improves the installation and upgrade experience for Plandex local mode users.

✨ Key Features

🔄 Smart Installation Updates

• Install script now detects and updates existing local server installations
• Prompts users to update rather than create duplicate installations
• Ensures server code is always on the latest version during setup

🔧 Comprehensive Upgrade Process

• CLI upgrades now automatically update the local Docker server
• Maintains synchronization between CLI and server versions
• Rebuilds and restarts Docker containers with latest code

🛠️ New Server Update Command

• Added plandex update-server command for manual server updates
• Provides detailed feedback during update process
• Handles Docker container management automatically

🐛 Bug Fixes

• Critical: Fixed nil pointer dereference in model fallback handling (from v2.2.5)
• Installation: Resolved issue where server runs old code after CLI upgrade
• Upgrade: Eliminated version mismatch between CLI and server components

🎯 User Experience Improvements

Before This Release

• CLI upgrades left server running old code
• Manual server updates required complex Docker commands
• Installation could create multiple server directories
• Version mismatches caused compatibility issues

After This Release

• ✅ CLI and server stay perfectly synchronized
• ✅ One-command server updates with plandex update-server
• ✅ Smart installation process handles existing setups
• ✅ Automatic Docker container management

🔧 Technical Details

Installation Process

curl -sL https://raw.githubusercontent.com/shipdocs/plandex/main/app/cli/install.sh | bash

• Detects existing ~/plandex-server installations
• Offers to update existing installations to latest code
• Rebuilds Docker containers with updated server code
• Maintains user configuration and data

Upgrade Process

• plandex automatically checks for updates and prompts user
• Updates CLI binary to latest version
• Pulls latest server code from repository
• Rebuilds and restarts Docker containers
• Verifies server health after restart

Manual Server Updates

plandex update-server

• Updates server code without touching CLI
• Handles git repository management
• Manages Docker container lifecycle
• Provides detailed progress feedback

🏗️ Architecture Improvements

• Unified Update Logic: Shared update functions between install script and CLI
• Error Handling: Comprehensive error checking and user feedback
• State Management: Preserves Docker container state during updates
• Version Synchronization: Ensures CLI and server compatibility

📋 Compatibility

• ✅ Backward Compatible: Works with existing installations
• ✅ Cross-Platform: Linux and macOS support
• ✅ Docker Integration: Seamless container management
• ✅ Git Integration: Handles repository updates safely

🚀 Enhanced Models (Continued from v2.2.4)

• Claude Sonnet 4.5: Latest Anthropic model with 1M token context
• NanoGPT Integration: Access to multiple providers through single API
• New Models: Gemini 2.5 Flash, DeepSeek V3, DeepSeek R1

Installation: curl -sL https://raw.githubusercontent.com/shipdocs/plandex/main/app/cli/install.sh | bash

Upgrade: Run plandex and follow the upgrade prompt, or download binaries manually

Server Update: plandex update-server (for existing local installations)

🔧 Critical Bug Fix Release

🐛 Fixed

🚨 Nil Pointer Dereference Fix

• Issue: Fixed critical runtime panic: "Invalid memory address or nil pointer dereference"
• Location: Model fallback handling in streaming operations
• Impact: Resolves crashes when analyzing codebases or executing model requests
• Root Cause: Missing BaseModelConfig in FallbackResult when no fallback was needed

🔍 Technical Details

Files Modified:

• app/shared/ai_models_errors.go: Set BaseModelConfig in fallback result
• app/server/model/client_stream.go: Added safety check for nil pointer access

Stack Trace Fixed:

🚨 Panic in execTellPlan
  → Runtime error
    → Invalid memory address or nil pointer dereference

✨ What's Included from v2.2.4

• Claude Sonnet 4.5: Upgraded with 1M token context
• Enhanced NanoGPT Support: Gemini 2.5 Flash, DeepSeek V3, DeepSeek R1
• Improved Performance: Better coding assistance and larger context handling

🛠️ Usage

# Install latest version with bug fix
curl -sL https://raw.githubusercontent.com/shipdocs/plandex/main/app/cli/install.sh | bash

# Use with NanoGPT
export NANOGPT_API_KEY='your-api-key'
plandex

🔄 Upgrade Recommendation

Highly recommended upgrade for all users experiencing:

• Crashes when analyzing codebases
• "Panic in execTellPlan" errors
• Model execution failures

This release maintains all the enhanced NanoGPT model support from v2.2.4 while fixing the critical stability issue.

🚀 Enhanced NanoGPT Model Support

✨ Model Updates

🧠 Claude Sonnet 4.5 Upgrade

• Upgraded: Claude Sonnet 4 → Claude Sonnet 4.5 (claude-sonnet-4-5-20250929)
• Context Length: Increased from 200K to 1M tokens
• Performance: Superior coding and agentic capabilities
• Benefits: Larger context window for complex projects

📦 New NanoGPT Provider Support

• Gemini 2.5 Flash: Added gemini-2.5-flash support
• DeepSeek V3: Added deepseek-chat support
• DeepSeek R1: Added deepseek-r1 reasoning model support

🎯 What's Improved

• Better Performance: Claude Sonnet 4.5 offers significantly improved coding assistance
• Expanded Access: More models available through single NanoGPT API key
• Larger Context: 1M token context for handling bigger codebases
• Latest Models: Access to cutting-edge reasoning models like DeepSeek R1

🛠️ Usage

# Install latest version
curl -sL https://raw.githubusercontent.com/shipdocs/plandex/main/app/cli/install.sh | bash

# Use with NanoGPT
export NANOGPT_API_KEY='your-api-key'
plandex

📋 Available Models via NanoGPT

• OpenAI: o3, o4-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano
• Anthropic: claude-sonnet-4.5 ⭐, claude-3.5-sonnet, claude-3.5-haiku
• Google: gemini-2.5-pro, gemini-2.5-flash ⭐, gemini-1.5-pro
• DeepSeek: deepseek-v3 ⭐, deepseek-r1 ⭐

⭐ = New or updated in this release

These updates ensure Plandex users have access to the most capable AI models available through NanoGPT's unified API.

🔧 NanoGPT Universal Provider Fix

Bug Fixes

• Universal Provider Logic: Fixed credential checking to recognize NanoGPT as universal provider
• CLI Binary: Updated CLI binary with latest universal provider support
• Install Script: Ensures both server and CLI have NanoGPT universal provider logic

What's Fixed

• NanoGPT now shows as "Quick option" when available for all publishers
• Single NanoGPT API key works for OpenAI, Anthropic, and Google models
• No more "missing provider" errors when using daily-driver model pack with NanoGPT

Usage

# Install with fixed NanoGPT support
curl -sL https://raw.githubusercontent.com/shipdocs/plandex/main/app/cli/install.sh | bash

# Set NanoGPT API key
export NANOGPT_API_KEY='your-api-key'
plandex

Should now work without credential errors! 🎉

🚀 NanoGPT Provider Integration

This release adds support for NanoGPT as a unified provider, allowing users to access multiple AI models through a single API key while maintaining Plandex's intelligent routing.

✨ New Features

🔌 NanoGPT Provider Support

• Added NanoGPT as a built-in provider with OpenAI-compatible endpoints
• Single API key access to models from OpenAI, Anthropic, Google, and more
• Seamless integration with existing model routing and fallback systems
• Interactive setup through the install script

📦 Supported Models

• OpenAI: o3, o4-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano
• Anthropic: claude-3.5-sonnet, claude-3.5-haiku

🛠️ Setup

Via Install Script:

curl -sL https://raw.githubusercontent.com/shipdocs/plandex/main/app/cli/install.sh | bash
# Choose option 2 for NanoGPT during provider setup

Manual Setup:

export NANOGPT_API_KEY='your-nanogpt-api-key'
plandex # start Plandex

📚 Documentation

• Updated model providers documentation with NanoGPT setup guide
• Added comprehensive test suite for provider integration
• Included failover and redundancy configuration examples

🎯 Benefits

• Simplified Management: One API key instead of multiple provider keys
• Smart Routing: Maintains Plandex's intelligent model selection
• Redundancy: Works alongside other providers for failover
• Easy Setup: Guided configuration through install script

For more details, see the model providers documentation.