This document tracks the major milestones and evolution of the Django Ergo project.
- The built-in routing tiers are now
small(Haiku 5.5, GPT-6 Luna),medium(Sonnet 5.5, GPT-6.1 Sol),large(Opus 5.5, GPT-6 Sol) and the newxlarge(Fable 5.1, GPT-6 Astra), each falling back to the pair below it. Effort is no longer tied to the tier: everything runs at medium unless a chat picks another level with the effort slider in its thread options (POST /api/sessions/<id>/effort).lowandhighstay as aliases ofsmallandlarge, so existingauto/lowchats and providers.yaml keys keep working. Haiku 5.5 pricing is added.
- Subscription limits were only recorded as a side effect of a Claude or Codex
CLI turn, and a partial event refreshed the row's timestamp while leaving
old windows in place, so idle providers showed stale 7-day values and no
Claude 5-hour window.
bots/usage_sync.pynow runsomp usage --redact --jsonevery 5 minutes (and on Refresh limits), stores each window with its ownobserved_at, and records the sync's health inUsageSync. - The Routing page is redesigned as Routing & capacity: sync health, a card per account with 5-hour and 7-day windows, rate ticks showing the elapsed share of each window, and explicit Unavailable/Stale states.
- Routing uses only reported subscription windows, preserving unknown utilization, reset times and status; Claude's Fable weekly allowance only gates Fable models.
- The Routing page separates pay-per-token providers from subscriptions and shows usage freshness instead of inventing 5-hour/weekly counters.
- Built-in low/medium/high tiers use configured current-catalog subscription
models, with Fable/Astra high-capability defaults and medium-model fallbacks.
providers.yamlsupports custom names and independent per-tier overrides.
- 8283455 - Initial commit
- dd0b5a8 - Cursor Dockerfile development environment setup
- addf06c - PostgreSQL configuration and test infrastructure
- 0aea1b9 - Cursor AI development standards and rules
- b5589a2 - Added
old-code-inspiration/with prototype models and workflow engine - 4ad3c4d - Created comprehensive planning documentation:
- Architecture analysis of prototype
- 24-week development roadmap
- Detailed TODO lists and questions
- Product specifications for example apps
- 1a6d8f7 - Refined product documentation for garden management and personal goals apps
- 25cab62 - Finalized v1.0 architectural decisions:
- MCP integration strategy (reusable tools for Django apps)
- Pluggable embedding system with OpenAI default
- Workflow engine with context serialization
- App-level permissions model
- b463f22 - Implemented core Django Ergo models and functionality:
- Migrated models from prototype (Workflow, Knowledgebase, Article, UserChat, ChatMessage)
- Built tool system with 6 knowledge base tools
- Created workflow engine with state persistence
- Enhanced Django admin interface
- Sample data management command
- caf4fa6 - Improved startup scripts with logging and debugging
- 940eee3 - Implemented dual-tier OpenAI testing with fixtures:
- Test mode for development without API costs
- Production mode with real OpenAI integration
- Comprehensive test coverage for tool system
- Core Models: All prototype models migrated and enhanced
- Tool System: 6 knowledge base tools with registry and validation
- Workflow Engine: Basic processing with state persistence and approval system
- Admin Interface: Full Django admin integration
- Development Environment: Complete Cursor setup with all dependencies
- Testing Infrastructure: Comprehensive test suite with fixtures
- PostgreSQL + pgvector: Production database setup
- OpenAI Integration: Real API integration (currently in test mode)
- Public APIs: REST/GraphQL API design and implementation
- v0.2.0: Production-ready embeddings and search
- v0.3.0: Public APIs and authentication
- v1.0.0: Complete MCP integration and example applications
Analyzed existing prototype to understand:
- Multi-tenant knowledge base architecture
- Hybrid search with PostgreSQL + pgvector
- Extensible tool system with function decorators
- Workflow engine with OpenAI integration
Migrated and enhanced core components:
- Django models with proper migrations
- Tool registry with automatic discovery
- Admin interface with custom views
- Development and testing infrastructure
Building towards production readiness:
- Real embedding generation and vector search
- OpenAI API integration with proper error handling
- Public APIs for external access
- Performance optimization and caching
- MCP Strategy: Focus on reusable tools for Django apps rather than monolithic platform
- Embedding System: Pluggable providers with OpenAI default and custom embedding support
- Workflows: Python-based with OpenAI context serialization for pause/resume
- Permissions: App-level rather than framework-level for maximum flexibility
- Knowledge Management: Flatfile export/import for agentic processing
Created comprehensive planning documents:
- Architecture analysis and decisions
- 24-week roadmap with phases
- Detailed TODO lists (299 items)
- 60+ questions for stakeholder input
- Product specifications for example apps
Streamlining documentation for maintainability:
- Consolidated redundant summary documents
- Cleaned up resolved questions
- Updated history with actual project evolution
- Focused on actively used planning documents
- leeward bound - Project lead and primary developer
- Cursor Background Agents - Architecture, planning, and implementation assistance
- Empty Django project structure
- Complete core model implementation
- Tool system with 6 knowledge base tools
- Workflow engine with state persistence
- Django admin integration
- Comprehensive test suite
- Development environment setup
This history is maintained as a living document and updated with each major milestone.