Claude Conda Manager

A dedicated Claude Code workspace for managing and optimizing Conda environments on Daniel's Ubuntu workstation.

Purpose

This repository serves as a Claude Code Space - a specialized local workspace where Claude assists with systematic Conda environment management rather than traditional code development. The goal is to maintain an efficient, well-organized collection of Conda environments optimized for AI/ML workflows on AMD ROCm hardware.

What is a Claude Code Space?

A Claude Code Space is a dedicated repository on your local machine used for specific administrative tasks and system management through Claude Code. Unlike traditional development repositories, these spaces are:

• Focused on local system administration and maintenance
• Used for organizing complex tasks that benefit from documentation and version control
• A persistent workspace where Claude can track configurations, decisions, and changes over time
• Version-controlled for accountability and rollback capabilities

Core Responsibilities

Environment Management

• Create and maintain baseline general-purpose environments with commonly-needed packages
• Build specialized environments for specific workflows (ML training, STT, video processing, etc.)
• Remove duplicate or broken environments
• Verify environment functionality and package compatibility

AMD ROCm Optimization

• Navigate complex ROCm dependency chains
• Ensure PyTorch and other frameworks are properly configured for AMD GPUs
• Maintain documentation of working ROCm + package combinations

Documentation & Best Practices

• Track environment configurations and their purposes
• Document successful package combinations and known issues
• Provide guidance on choosing between Conda, pip, UV, and other package managers
• Suggest efficient environment-switching workflows

User Context

Hardware

• CPU: Intel Core i7-12700F (12 cores, 20 threads)
• GPU: AMD Radeon RX 7700 XT / 7800 XT (Navi 32, gfx1101)
• Memory: 64 GB RAM
• Storage: ~4.6 TB total available

Primary Use Cases

• AI/ML model development and fine-tuning
• Speech-to-text processing
• Video editing and generation
• Data analysis and visualization
• General automation and scripting

Conda Setup

• Miniconda3: ~/miniconda3
• Anaconda3: ~/anaconda3
• Both installations available for different purposes

Directory Structure

Claude-Conda-Manager/
├── README.md                 # This file
├── CLAUDE.md                 # Instructions for Claude Code
├── envs/                     # Environment documentation and configs
├── docs/                     # Additional documentation
├── templates/                # Environment templates and recipes
├── logs/                     # Operation logs and verification results
└── backups/                  # Environment export files

Key Principles

1. Minimize Complexity: Reuse existing environments before creating new ones
2. ROCm First: Prioritize ROCm compatibility for GPU-accelerated workflows
3. Document Everything: Track what works, what doesn't, and why
4. Test Before Deploy: Verify environments before declaring them ready
5. Version Control: Keep environment configs and documentation in Git

Typical Workflows

Creating a Base Environment

Claude evaluates your typical workloads and creates general-purpose environments with pre-installed heavy packages (PyTorch, TensorFlow, common ML libraries).

Specialized Environment Setup

When you need specific capabilities (e.g., LLM fine-tuning, STT model training), Claude builds targeted environments with minimal redundancy.

Environment Audit

Periodic reviews to identify duplicate, broken, or unused environments for cleanup.

Dependency Resolution

When ROCm or package conflicts arise, Claude investigates and documents solutions.

Package Manager Guidance

• Conda: Use for ML/AI packages, complex dependencies, and ROCm-specific builds
• pip: Use within Conda environments for packages not available through Conda
• UV: Use for lightweight, quick project venvs outside of Conda
• pipx: Use for isolated CLI tool installations

Getting Started

This workspace is already configured. Simply open Claude Code in this directory and request environment management tasks:

• "Audit my current Conda environments"
• "Create a general ML environment with ROCm support"
• "Check if I have any duplicate environments"
• "Set up an environment for LLM fine-tuning"
• "Document my working PyTorch + ROCm configuration"

Version Control

This repository is version-controlled to track:

• Environment configurations over time
• Decisions about environment organization
• Working package combinations
• Known issues and solutions

Changes are pushed to GitHub for backup and historical reference.

Related Resources

• Conda Documentation: https://docs.conda.io/
• ROCm Documentation: https://rocm.docs.amd.com/
• PyTorch ROCm: https://pytorch.org/get-started/locally/

---

Repository Type: Claude Code Space Primary User: Daniel Rosehill Location: ~/repos/github/Claude-Conda-Manager Last Updated: 2025-10-28