An AI-powered CLI tool for scaffolding projects with built-in documentation, learning systems, and product-loop workflows designed for use with Claude Code and Cursor AI.
Learned from real practice - Built on insights from 5+ billion tokens of development experience and commercial products. Extracts best practices from chat history, documentation patterns, and proven workflows.
pip install up-cli# Create new project
up new my-project
# Or initialize in existing project
cd existing-project
up init
# Start AI-assisted development with safety rails
up save # Checkpoint before AI work
up start # Run AI product loop
up diff # Review changes
up review # AI adversarial review
# Check system health
up status
# Live dashboard
up dashboard| Command | Description |
|---|---|
up new <name> |
Create a new project with full scaffolding |
up new <name> --template <type> |
Create project from specific template |
up init |
Initialize up systems (auto-installs git hooks, builds memory) |
up init --ai claude |
Initialize for Claude Code only |
up init --ai cursor |
Initialize for Cursor AI only |
up init --systems docs,learn |
Initialize specific systems only |
up start |
Start the product loop |
up start --resume |
Resume from last checkpoint |
up start --dry-run |
Preview mode without changes |
up start --parallel |
Parallel multi-agent execution |
up save |
Create checkpoint before AI work |
up reset |
Restore to checkpoint |
up diff |
Review AI changes |
up review |
AI adversarial code review |
up agent spawn <name> |
Create agent worktree |
up agent status |
List all agents |
up agent merge <name> |
Squash and merge |
up bisect |
Find bug-introducing commit |
up provenance list |
View AI operation history |
up branch status |
Show branch hierarchy |
up status |
Show health of all systems |
up dashboard |
Live interactive health dashboard |
up sync |
Sync all systems (memory, docs) |
up hooks |
Install/manage git hooks for auto-sync |
up learn |
Auto-improve project with AI (requires vision map) |
up learn "topic" |
Learn about specific topic with AI research |
up learn "file.md" |
Analyze file with AI (auto-fallback to basic) |
up learn "path" |
Compare and learn from another project |
up learn analyze |
Analyze all research files with AI + progress bar |
up learn plan |
Generate improvement PRD |
up learn --no-ai |
Disable AI (faster, basic extraction only) |
up memory search <query> |
Semantic search in memory |
up memory sync |
Index git commits and files |
up memory branch |
Show branch-specific knowledge |
up memory record |
Record learnings/decisions/errors |
up summarize |
Summarize AI conversation history |
Create projects with pre-configured tech stacks:
# FastAPI backend with SQLAlchemy
up new my-api --template fastapi
# Next.js frontend with TypeScript
up new my-app --template nextjs
# Python library with packaging
up new my-lib --template python-lib
# Minimal structure
up new my-project --template minimal
# Full setup with MCP
up new my-project --template full| Template | Description |
|---|---|
minimal |
Basic structure with docs |
standard |
Full up systems (default) |
full |
Everything including MCP server |
fastapi |
FastAPI + SQLAlchemy + pytest |
nextjs |
Next.js 14 + TypeScript + Tailwind |
python-lib |
Python library with pyproject.toml |
# Create a new project with all systems
up new my-saas-app
# Create with a specific template
up new my-api --template fastapicd my-existing-project
# Full initialization
up init
# Claude Code focused setup
up init --ai claude
# Only add docs and learn systems
up init --systems docs,learn# Quick status check
up status
# Live dashboard (updates every 5 seconds)
up dashboard
# JSON output for scripting
up status --jsonAll learn commands use Claude/Cursor AI by default with automatic fallback.
# Self-improvement analysis (requires configured vision map)
up learn
# Learn about a specific topic (AI-powered research)
up learn "caching strategies"
up learn "authentication"
up learn "testing best practices"
# Learn from another project's design
up learn "../other-project"
up learn "~/projects/reference-app"
# Learn from a file (AI-powered analysis)
up learn "docs/architecture.md"
up learn "guide.txt"
# Analyze all research files with AI + progress bar
up learn analyze
# Auto-analyze without vision map requirement
up learn auto
# Check learning system status
up learn status
# Generate a PRD from analysis
up learn plan
# Disable AI for faster basic extraction
up learn --no-ai "docs/guide.md"The learn system uses AI by default:
- Self-improvement (
up learn): Analyzes current project with AI insights. Requires a configureddocs/roadmap/vision/PRODUCT_VISION.md. - Topic learning (
up learn "topic"): AI-powered research for specific topics based on your project's tech stack. - File learning (
up learn "file.md"): AI analysis of files with automatic fallback to basic extraction. - Batch analysis (
up learn analyze): Process all research files with AI and tqdm progress bar. - Basic mode (
--no-ai): Skip AI for faster basic regex extraction.
# Start the product loop
up start
# Resume from checkpoint
up start --resume
# Preview what would happen
up start --dry-run
# Start with specific task
up start --task US-003
# Use custom PRD file
up start --prd path/to/prd.json# Summarize Cursor chat history
up summarize
# Export as JSON
up summarize --format json --output summary.json
# Filter by project
up summarize --project myprojectComprehensive documentation structure:
docs/
├── CONTEXT.md # AI reads first
├── INDEX.md # Quick reference
├── roadmap/ # Strategic planning
│ ├── vision/ # Product vision
│ └── phases/ # Phase roadmaps
├── architecture/ # System design
├── features/ # Feature specs
├── changelog/ # Progress tracking
├── handoff/ # Session continuity
├── decisions/ # ADRs
└── learnings/ # Patterns discovered
Research and improvement pipeline:
RESEARCH → ANALYZE → COMPARE → PLAN → IMPLEMENT
Three learning modes:
| Mode | Command | Description |
|---|---|---|
| Self-improvement | up learn |
Analyze and improve current project (requires vision map) |
| Topic learning | up learn "topic" |
Create research file for specific topic |
| External learning | up learn "path" |
Learn from project directory or file |
Supported file types for learning:
- Documentation:
.md,.markdown,.txt,.rst - Python:
.py(extracts patterns, classes, functions) - JavaScript/TypeScript:
.js,.ts,.tsx,.jsx - Config:
.json,.yaml,.yml,.toml
Additional commands:
up learn auto- Analyze without vision map requirementup learn analyze- Extract patterns from research filesup learn plan- Generate improvement PRDup learn status- Show learning system status
Storage:
.claude/skills/learning-system/
├── project_profile.json # Current project analysis
├── research/ # Topic research files
├── external_learnings/ # Learnings from other projects
├── file_learnings/ # Learnings from individual files
├── insights/ # Extracted patterns
└── prd.json # Generated improvement plan
Autonomous development with safety guardrails:
| Principle | Implementation |
|---|---|
| Stable | Graceful degradation, fallback modes |
| Efficient | Token budgets, incremental testing |
| Safe | Input validation, path whitelisting |
| Reliable | Timeouts, idempotency, rollback |
| Cost-effective | Early termination, ROI threshold |
Features:
- Circuit breaker (max 3 failures)
- Checkpoint/rollback
- Health checks
- Budget limits
Tracks AI context window usage:
- Estimates token usage per file/message
- Warns at 80% capacity
- Suggests handoff at 90%
- Persists across sessions
Persistent knowledge that survives across sessions:
# Search for relevant knowledge
up memory search "authentication"
# Sync git commits and files to memory
up memory sync
# Record learnings and decisions
up memory record --learning "Use dataclasses for configs"
up memory record --decision "Chose PostgreSQL for ACID compliance"
# View branch-specific knowledge
up memory branch
up memory branch feature-x --compare mainFeatures:
- Semantic search using ChromaDB (local embeddings, no API required)
- Branch/commit-aware - knowledge tagged with git context
- Auto-indexing - git hooks sync commits automatically
- Cross-session persistence - remembers learnings, decisions, errors
Storage:
.up/
└── memory/
└── chroma/ # ChromaDB vector database
Model Context Protocol integration:
.mcp/
├── config.json # Server configuration
├── tools/ # Custom tool definitions
└── README.md # Usage guide
When you run up init, git hooks are automatically installed:
# Git hooks auto-installed by up init
.git/hooks/
├── post-commit # Auto-indexes commits to memory
└── post-checkout # Updates context on branch switchThis means your knowledge is captured automatically:
- Every
git commitis indexed to memory - Branch switches update context
- No manual sync required for commits
| File | Purpose |
|---|---|
CLAUDE.md |
Claude Code instructions |
.cursorrules |
Cursor AI rules |
.cursor/rules/*.md |
File-specific rules |
.claude/context_budget.json |
Context tracking |
.up/memory/ |
Long-term memory storage |
Generated rules for different file types:
main.md- General project rulespython.md- Python standardstypescript.md- TypeScript standardsdocs.md- Documentation standardstests.md- Testing standards
Design for AI collaboration, not just human readability.
- Context-aware scaffolding - Project structures optimized for AI agents to navigate and understand quickly
- Explicit over implicit - Clear file naming, directory structures, and documentation that AI can parse without ambiguity
- Prompt-friendly patterns - Code and docs written to be easily referenced in AI conversations
- Tool integration - Native support for Claude Code skills and Cursor AI rules
Documentation is the source of truth, not an afterthought.
- Docs-first workflow - Write documentation before implementation to clarify intent
- Living documentation - Docs evolve with the codebase through automated learning systems
- Knowledge extraction -
/learncommands analyze patterns and generate insights from real usage - Structured knowledge - Vision, roadmaps, and changelogs in predictable locations for AI and human consumption
Autonomous development with safety guardrails.
- Circuit breaker protection - Max 3 consecutive failures before stopping to prevent runaway loops
- Checkpoint/rollback - Save state before risky operations, restore on failure
- Health checks - Validate system state between iterations
- Budget limits - Token and time constraints to prevent unbounded execution
- Human-in-the-loop - Critical decisions require explicit approval
| Practice | Description |
|---|---|
| Incremental delivery | Ship small, working increments over big-bang releases |
| Fail fast, recover faster | Detect issues early, rollback automatically |
| Observable by default | Logging, metrics, and state visible to both AI and humans |
| Convention over configuration | Sensible defaults that work out of the box |
# Install for development
pip install -e ".[dev]"
# Run tests
pytest
# Lint
ruff check src/
# Type check
mypy src/up-cli/
├── src/up/
│ ├── cli.py # Main CLI
│ ├── context.py # Context budget management
│ ├── memory.py # Long-term memory (ChromaDB)
│ ├── events.py # Event-driven integration
│ ├── summarizer.py # Conversation analysis
│ ├── commands/ # CLI commands
│ │ ├── init.py # Initialize project
│ │ ├── new.py # Create new project
│ │ ├── status.py # System health
│ │ ├── dashboard.py # Live monitoring
│ │ ├── learn.py # Learning system
│ │ ├── memory.py # Memory commands
│ │ ├── sync.py # Sync & hooks
│ │ ├── start.py # Product loop
│ │ └── summarize.py # Conversation summary
│ └── templates/ # Scaffolding templates
│ ├── config/ # CLAUDE.md, .cursor/rules
│ ├── docs/ # Documentation system
│ ├── learn/ # Learning system
│ ├── loop/ # Product loop
│ ├── mcp/ # MCP server
│ └── projects/ # Project templates
├── scripts/ # Utility scripts
│ ├── export_claude_history.py
│ └── export_cursor_history.py
├── docs/ # Documentation
│ ├── architecture/ # System architecture
│ └── guides/ # Usage guides
└── skills/ # Reference skills
MIT