Skip to content
BSWEN
Blog

oh-my-pi vs Claude Code: Which AI Coding Harness Should You Use?

Cowrie
Cowrie
Dev @ Bswen

Developer workspace with terminal

Choosing an AI coding harness used to be simple. You picked Claude Code if you used Claude models. Now the landscape has changed. I’ve spent weeks comparing two terminal-native AI coding agents: Claude Code (Anthropic’s official tool) and oh-my-pi (an open-source alternative). Here’s what I found.

The Core Question

Both tools connect AI models to your code, files, and terminal. Both support MCP (Model Context Protocol), LSP integration, and subagents. But they differ in philosophy and target users.

The real question isn’t which is “better” - it’s which matches your workflow and expertise level.

Quick Answer

Claude Code is the safer default for most developers. It has mature ecosystem, official documentation, and works out of the box.

oh-my-pi wins if you need role-based model routing (GLM-5 for planning, Qwen for execution), hash-anchored edits for concurrent work, or heavy customization.

Feature Comparison

Key differences
| Feature              | Claude Code           | oh-my-pi              |
|----------------------|----------------------|-----------------------|
| Developer            | Anthropic (official) | Open-source (can1357) |
| Model Support        | Claude + routing     | 50+ providers unified |
| Role-Based Routing   | Via skills/config    | Built-in              |
| Hash-Anchored Edits  | No (line-based)      | Yes                   |
| MCP Protocol         | Native               | Integrated            |
| IPython Kernel       | No                   | Yes (persistent)      |
| Context Compaction   | Manual (/compact)    | Automatic             |
| Customization Level  | Moderate             | High                  |

What Makes oh-my-pi Different

Role-Based Model Routing

oh-my-pi has built-in role-based routing. You configure which model handles which task type:

oh-my-pi role configuration
roles:
  slow:
    model: glm-5.1
    description: Deep reasoning for planning
  plan:
    model: glm-5.1
    description: Architecture and strategy
  default:
    model: qwen-3.5
    description: Fast execution

This means expensive reasoning models handle planning, while cheaper fast models handle routine edits. The cost savings are real - I’ve seen my API costs drop 40% since switching planning tasks to GLM-5 and execution to Qwen.

Claude Code can approximate this via skills with model overrides, but it requires more setup:

Claude Code skill for deep planning
# .claude/skills/planning-skill/SKILL.md
---
name: deep-plan
description: Research and architecture planning
model: opus
context: fork
allowed-tools: Read, Grep, Glob
---


You are a planning agent. Think deeply before acting.

Hash-Anchored Edits

This is oh-my-pi’s technical edge. When editing files, it uses content-hash-based addressing instead of line numbers.

Why this matters: When multiple agents edit the same file concurrently, line numbers drift. Agent A edits line 50, shifting everything down. Agent B’s edit targeting “line 60” now hits wrong content.

Hash-anchored edits reference content by hash:

Hashline addressing example
// Hashline: @abc123def456
const originalLine = "target content";

The agent finds the content by hash, not position. Line drift doesn’t break subsequent edits.

I ran into this problem with Claude Code when spawning parallel subagents. One agent’s edits shifted line numbers, causing another’s edits to fail. With oh-my-pi, concurrent edits worked reliably.

What Makes Claude Code Better

Ecosystem and Documentation

Claude Code has official Anthropic support. Regular updates, comprehensive docs, active community.

The skills framework is mature - auto-invocation, context-aware triggers, inheritance from CLAUDE.md hierarchy.

Hooks system covers PreToolUse, PostToolUse, and Stop events:

Claude Code hook configuration
// .claude/settings.json
{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          { "type": "command", "command": "pnpm lint:fast" }
        ]
      }
    ]
  }
}

oh-my-pi has equivalent features but requires more knowledge to configure effectively.

MCP Integration

Claude Code has first-class MCP support. GitHub MCP, database MCP, file system MCP - they all integrate cleanly.

I use the GitHub MCP daily for PR creation and issue tracking. The integration feels native, not bolted on.

Common Misconceptions

I made these mistakes initially:

“Claude Code only works with Claude models” - Wrong. You can route other models through Claude Code, but the setup is less elegant than oh-my-pi’s native role system.

“oh-my-pi is simpler” - Wrong. It’s actually more complex under the hood (Rust + TypeScript). “Simpler” refers to terminal-first philosophy, not implementation.

“More features = better” - Wrong. Claude Code has more features, but oh-my-pi’s focused features (role routing, hash edits) solve specific problems better.

Decision Guide

Choose Claude Code if:

  • You want minimal setup
  • You’re already a Claude Pro subscriber
  • You value official documentation
  • You use MCP servers heavily
  • You’re new to AI coding agents

Choose oh-my-pi if:

  • You want role-based model routing
  • You need concurrent multi-agent editing
  • You have harness engineering experience
  • You use non-Claude models primarily
  • You need persistent IPython for data science

The hybrid approach:

Some developers run both. Claude Code for everyday coding with its ecosystem. oh-my-pi for specific tasks needing role routing or hash-anchored edits.

One Reddit user noted: “Even my Claude Code is highly modified.” Power users customize Claude Code extensively. oh-my-pi simply makes customization more accessible.

Bottom Line

Winner by category
| Category            | Winner                              |
|--------------------|--------------------------------------|
| Ease of Use        | Claude Code                          |
| Cost Optimization  | oh-my-pi (role routing)              |
| Ecosystem          | Claude Code                          |
| Customization      | oh-my-pi                             |
| Concurrent Editing | oh-my-pi (hash-anchored)             |
| Non-Claude Models  | oh-my-pi                             |
| Documentation      | Claude Code                          |

Your harness engineering knowledge determines which tool serves you better. If you’re new to AI coding agents, start with Claude Code. If you understand harness internals and want specific advantages, oh-my-pi gives you more control.

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!

Comments

Tags: