Skip to content
Β 
Β 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

210 Commits
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

Claudex

The Claude Experience you wish you had.

Important: Claude Pro/Max/Team subscription required

Claudex requires Claude Code, which is only available with a Claude Pro, Max, or Team subscription. The free Claude tier does not include Claude Code access.

Features

πŸ—‚οΈ Persistent Sessions

Every task starts with a session β€” a folder that accumulates everything Claude produces:

.claudex/sessions/
└── api-refactor-abc123/
    β”œβ”€β”€ session-overview.md    ← Auto-maintained status & index
    β”œβ”€β”€ feature-description.md ← Manually added from Jira, Linear, etc.
    β”œβ”€β”€ research-findings.md   ← Research artifacts
    β”œβ”€β”€ execution-plan.md      ← Architecture decisions
    └── ...                    ← Your custom docs

Why it matters: Claude's context window fills up. When you clear it, Claude normally forgets everything. With claudex, the session folder persists β€” Claude reads session-overview.md on startup and catches up in seconds.

Session modes:

  • Resume β€” Continue where you left off with full claude's conversation history
  • Fresh memory β€” Clear claude's context window, keep all docs (Claude catches up via overview)
  • Fork β€” Branch into a new task while cloning all the docs

πŸ“ Auto-Documentation

A background agent silently maintains session-overview.md as you workβ€”no manual note-taking:

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  You work normally                                                    β”‚
β”‚       ↓                                                               β”‚
β”‚  Every few messages, claudex updates the session-overview.md document β”‚
β”‚       ↓                                                               β”‚
β”‚  Clear claude's context window                                        β”‚
β”‚       ↓                                                               β”‚
β”‚  Leverage full claude's potential                                     β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Example of auto-maintained session-overview.md:

# Session: API Refactor

## Status
Phase 2 in progress - Authentication endpoints complete

## Key Decisions
- JWT over session cookies (see research-auth.md)
- Rate limiting at gateway level

## Documents
- [research-auth.md](./research-auth.md) β€” Auth strategy analysis
- [execution-plan.md](./execution-plan.md) β€” Implementation phases

Pick up any session instantlyβ€”even weeks later. Claude reads the overview, follows the pointers, and catches up in seconds.

πŸ“š Auto-Updating Index Files

Keep your codebase documentation up-to-date automatically. On first run in a git repo, claudex offers to install a post-commit hook:

πŸ“ Enable auto-docs update after git commits? [y/n/never]:

When enabled, after each commit:

  1. Detects which files changed
  2. Identifies affected index.md files
  3. Spawns Claude (Haiku) to intelligently update them
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  git commit                                                      β”‚
β”‚       ↓                                                          β”‚
β”‚  Post-commit hook triggers claudex --update-docs                 β”‚
β”‚       ↓                                                          β”‚
β”‚  Claude reads existing index.md, explores surrounding context    β”‚
β”‚       ↓                                                          β”‚
β”‚  Makes thoughtful updates to keep docs relevant                  β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Manual trigger: claudex --update-docs

Skip for a commit: CLAUDEX_SKIP_DOCS=1 git commit -m "quick fix"

πŸ€– Parallel Agent Orchestration

A team-lead agent coordinates specialists through a structured workflow:

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  You describe what you need                         β”‚
β”‚       ↓                                             β”‚
β”‚  Explore agent investigates codebase & docs         β”‚
β”‚       ↓                                             β”‚
β”‚  Plan agent creates execution plan with phases      β”‚
β”‚       ↓                                             β”‚
β”‚  Engineers execute in parallel:                     β”‚
β”‚       β”œβ”€β”€ Track A: Auth service                     β”‚
β”‚       β”œβ”€β”€ Track B: API endpoints                    β”‚
β”‚       └── Track C: Database migrations              β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Parallel Agent Execution

Work gets broken into independent tracks. Multiple engineers execute simultaneously: divide-and-conquer.

Prerequisites

  • Claude Pro, Max, or Team subscription β€” Required for Claude Code access
  • Claude Code CLI β€” Install via npm install -g @anthropic-ai/claude-code
  • Node.js 14+ β€” For npm installation
  • Go 1.21+ β€” Only needed if building from source

Recommended MCPs

On first run, claudex will prompt you to install these recommended MCPs for the best experience:

MCP Description
Sequential Thinking Structured reasoning through complex problems step-by-step
Context7 Up-to-date documentation lookup for libraries and frameworks

You can also configure them manually anytime with claudex --setup-mcp or make install-mcp.

Installation

npm (Recommended)

npm install -g @claudex/cli

This works on macOS and Linux without needing to clone the repository.

From Source

git clone https://github.com/mgonzalezbaile/claudex.git
cd claudex
make install

Add to your shell config if needed:

export PATH="$HOME/.local/bin:$PATH"

Usage

Navigate to your project directory and run:

cd /path/to/your/project
claudex

On first run, claudex creates a .claude folder with agent profiles and hooks. If a .claude folder already exists, files are merged (use --no-overwrite to preserve your existing files).

The TUI will guide you through:

  1. Session selection (new, ephemeral, or existing)
  2. Profile selection (choose agent type)
  3. Launch Claude with your selections

Keyboard Controls

  • ↑/↓ - Navigate
  • Enter - Select
  • / - Fuzzy search
  • q or Ctrl+C - Quit

Agent Profiles

Claudex includes specialized agent profiles:

Profile Purpose
team-lead Strategic planning and orchestration
principal-engineer-{stack} Implementation (TypeScript, Python, Go, PHP)
prompt-engineer Prompt design and optimization

Profiles are automatically assembled based on your project's technology stack.

Configuration

Claudex stores its artifacts in a .claudex/ folder in your project root:

.claudex/
β”œβ”€β”€ config.toml      # Configuration file (auto-created)
β”œβ”€β”€ sessions/        # Session data
β”œβ”€β”€ logs/            # Log files
└── preferences.json # User preferences

Customizing Behavior

Edit .claudex/config.toml to customize behavior:

# Documentation files always loaded into context
doc = ["docs/index.md"]

# Preserve existing .claude files during setup
no_overwrite = true

[features]
# Auto-documentation during session (default: true)
autodoc_session_progress = true

# Auto-documentation on session end (default: true)
autodoc_session_end = true

# Tool executions between doc updates (default: 5)
autodoc_frequency = 5

Environment variables override config values: CLAUDEX_AUTODOC_SESSION_PROGRESS, CLAUDEX_AUTODOC_SESSION_END, CLAUDEX_AUTODOC_FREQUENCY.

Tip: Keep doc files lightweightβ€”they're passed to every agent. Use an index with brief descriptions and pointers:

# Project Documentation Index

## Product
- [docs/product-overview.md](docs/product-overview.md) β€” Business goals, user personas, success metrics

## Technology
- [docs/architecture.md](docs/architecture.md) β€” System design, service boundaries, data flow
- [docs/tech-stack.md](docs/tech-stack.md) β€” Languages, frameworks, infrastructure choices

## Development
- [docs/coding-standards.md](docs/coding-standards.md) β€” Style guide, patterns, conventions
- [docs/testing-strategy.md](docs/testing-strategy.md) β€” Test types, coverage requirements

Claude reads the index, understands what's available, and loads detailed docs on demandβ€”saving context for actual work.

License

MIT License - see LICENSE for details.

Credits

Built with:

About

The Claude Experience you wish you had

Resources

Code of conduct

Contributing

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages