Compare with Oh My OpenCode

[ret1]

Hi everyone,

I’d like to understand how OpenAgentsControl compares to Oh My OpenCode. Are there clear advantages of one over the other, or are they designed for different purposes?

Are there specific use cases where one should be preferred over the other?

Thanks in advance for your insights.

Best regards,
Thomas

[ramonskie]

im interested in this as well.
i just removed "oh my opencode". as with every release it seems to burn more and more tokens

[ret1]

hi @ramonskie !

thanks for your insights. have you already tried one of the recent releases of OpenAgentsControl? can you tell me if there is any difference in produced code quality?

oh my opencode 3 released recently, but after just a few minutes i hit my 4 hour limit with Claude Pro which is really annoying.

Bg,
Thomas

[ramonskie]

i have not uses openagents yet. but it same way more leaner that oh my opencode.
as you are more in controll.

oh my opencode seems heavily opinionated and. and it just fires off at will.
the first few versions where okay. but it got more bloated over time.

what i would like to see is that opus is only used when actually needed
but it seems like oh my opencode and openagents heavily rely on opus for controlling the right sub agents.

but i hope @darrenhinde could shine a light on this.

[darrenhinde]

Hey all,
Thanks for tagging me in this as I didn't see these messages.

So OpenAgentsControl (AOC) has a few guided principles that I follow when building this:

• Repeatable behaviour is paramount - we want repeatable results so we can rely on our tool
• This is made for teams - which means it needs to be able to be customised
• We care about context very much - making sure the right agent has the correct context without assuming it does
• How do we make this work in complex workspaces - with moving parts & different business decisions
• Be as token efficient as possible - less tokens in the window = better performance
• Have approval gates before iteration - results should be human guided. AI messes up intention, this is a quality gate must-have

I have purposely stayed away from the plugin systems as I wanted to make the framework first and really drill down into getting the most efficient system possible.
I use this everyday so I am constantly finding ways to improve it and get more out of it.

I wanted this to be used for any model so having access to the prompts is important to customise the behaviour.

---

In my opinion I look at oh-my-opencode as the beginner set it and forget it plugin. It does so much and blows your mind and your tokens but getting repeatable behaviour seems tough. You are at mercy of the plugin, if it updates well that means your prompt and behaviour updates... you have no say be at its mercy.

OpenAgentControl is that you have control on what you do, everything is accessible & you can change it as you will, if you want to update you can choose which files to update. Built around approval and efficient context methods.

I am actually making a video on how best to use this, as its getting closer to where I want it, last things missing are parallel agent control and abilities.

I made a comparison below and hopefully this sheds some light.

---

Comprehensive Feature Comparison: OpenAgentsControl vs Oh My OpenCode

Aspect	OpenAgentsControl	Oh My OpenCode
Philosophy	Plan-first, approval-based, team-ready	Batteries-included, autonomous orchestration
Primary Use Case	Production code matching YOUR patterns	Rapid iteration with multi-agent orchestration
🎯 Core Behavior
Agent Prompts	✅ Fully editable markdown files	❌ Baked into plugin code
Behavior Customization	✅ Complete control via prompt editing	⚠️ Limited (via prompt_append only)
Model Variants	✅ Separate prompt files per model	❌ Same prompt, different models
Team Adaptation	✅ Edit prompts to match team standards	⚠️ Append instructions only
Workflow Control	✅ Define your own workflows	❌ Fixed (ultrawork/Prometheus modes)
Extensibility	✅ Edit anything (prompts, context, agents)	⚠️ Fork repo to change core behavior
⚙️ Configuration
Settings Config	✅ Model, temp, basic options	✅✅ Model, temp, permissions, thinking, reasoning, tools, etc.
Per-Agent Config	✅ Model, temperature, prompt	✅✅ 15+ options per agent (see below)
Permission System	⚠️ Via agent configs	✅ Fine-grained (edit/bash/webfetch/doom_loop/external_directory)
Disable Components	✅ Agents, commands	✅✅ Agents, skills, hooks, commands, MCPs
Config Complexity	⭐⭐ Low-Medium	⭐⭐⭐⭐⭐ High (extensive options)
🤖 Agent System
Main Agents	3 (OpenAgent, OpenCoder, SystemBuilder)	9 (Sisyphus, Prometheus, Metis, Momus, Atlas, Oracle, Librarian, Explore, Multimodal-Looker)
Subagents	10+ (task-manager, coder, tester, reviewer, etc.)	Built into main agents
Category System	❌ (uses context files instead)	✅ 7 built-in categories (visual-engineering, ultrabrain, artistry, quick, unspecified-low/high, writing)
Custom Categories	❌	✅ Define your own with model/temp/prompt
Agent Specialization	✅ Via context loading	✅ Via category delegation
🔄 Execution & Control
Approval Gates	✅✅ Mandatory before ANY write/edit/bash	⚠️ Optional (ultrawork bypasses, Prometheus requires)
Auto-Fix on Error	❌ Never (always report first)	✅ Yes (in ultrawork mode)
Stop on Failure	✅ Always (request approval to continue)	⚠️ Continues (Ralph Loop)
Incremental Execution	✅ One step at a time, validate each	⚠️ Continuous until completion
Loop Until Done	⚠️ Limited (OpenCode default behavior)	✅✅ Ralph Loop + ulw-loop (keeps going until DONE)
User Involvement	⭐⭐⭐⭐⭐ High (approve each stage)	⭐⭐ Low (ultrawork) or ⭐⭐⭐⭐ High (Prometheus)
⚡ Parallelization & Background
Parallel Execution	⚠️ Limited (OpenCode default only)	✅✅ Full multi-agent orchestration
Background Tasks	⚠️ OpenCode default behavior	✅✅ delegate_task(run_in_background=true)
Concurrent Agents	⚠️ Not configurable	✅ Configurable per provider/model
Visual Multi-Agent	⚠️ Limited (OpenCode default only)	✅ Tmux integration (see agents in separate panes)
Background Concurrency Control	❌	✅ defaultConcurrency, providerConcurrency, modelConcurrency
💰 Token Efficiency
Design Principle	✅✅ MVI (Minimal Viable Information)	⚠️ Multi-agent orchestration (higher usage)
Context Overhead	✅✅ 80% reduction via MVI	⚠️ Higher (but parallelized)
Navigation Files	✅ ~200-300 tokens	⚠️ Full agent prompts
File Size Limits	✅ Enforced (100-200 lines max)	❌ No limits
Token Bloat Prevention	✅ Built-in (MVI enforced)	⚠️ Manual (configure cheap models)
Cost Optimization	✅✅ Token-efficient by design	⚠️ Requires configuration
📚 Context System
Context Loading	✅✅ Automatic via ContextScout	⚠️ Manual (AGENTS.md, rules injection)
YOUR Patterns	✅✅ Loaded automatically from context files	⚠️ Must configure per agent via prompt_append
Context Structure	✅ MVI files (~200 lines max)	⚠️ Full agent prompts (can be verbose)
Context Discovery	✅ ContextScout finds relevant files	⚠️ Librarian/Explore agents search
Standards Enforcement	✅✅ Loaded before ANY code work	⚠️ Via agent prompts (not enforced)
Project-Specific Patterns	✅ .opencode/context/project/project-context.md	⚠️ prompt_append per agent
✅ Quality & Validation
Code Quality Gates	✅✅ Load code-quality.md BEFORE code work	⚠️ Via agent prompts
Incremental Validation	✅ Type check, lint, test after each step	⚠️ Continuous (via LSP, diagnostics)
Testing	✅ TestEngineer subagent	⚠️ Manual or via skills
Code Review	✅ CodeReviewer subagent	⚠️ Oracle agent (read-only)
Comment Enforcement	⚠️ Manual	✅ Comment Checker hook
Refactoring Tools	✅ Edit tool + validation	✅✅ LSP rename + AST-grep
🎨 Advanced Features
LSP Integration	✅ Basic (diagnostics, symbols)	✅✅ Full (rename, references, code actions, custom servers)
AST-Grep	❌	✅ AST-aware code search/replace
Browser Automation	❌	✅ 2 providers (playwright, agent-browser)
Interactive Terminal	❌	✅ Tmux-based (vim, htop, pudb)
Session Tools	❌	✅ List, read, search, analyze sessions
Git Integration	✅ /commit command	✅✅ Git Master skill (atomic commits, auto-split, rebase, history search)
🔌 Extensibility
Skills System	✅ Built-in skills	✅✅ Built-in + custom + remote loading
Hooks System	⚠️ Limited	✅✅ 25+ hooks, all configurable
MCPs	✅ Built-in (websearch, context7, grep_app)	✅✅ Built-in + skill-embedded + OAuth
Commands	✅ Slash commands	✅✅ Slash commands + custom
Claude Code Compat	✅ Full compatibility layer	✅ Full compatibility layer
Custom Agents	✅ Easy (markdown files)	⚠️ Requires forking repo
🎛️ Hooks & Automation
Total Hooks	~10	25+
Context Injection	✅ AGENTS.md, README.md	✅ AGENTS.md, README.md, rules
Todo Enforcement	✅ Basic	✅✅ Todo Continuation Enforcer
Comment Checking	❌	✅ Comment Checker
Auto-Resume	❌	✅ Experimental
Dynamic Context Pruning	❌	✅ Experimental
Truncation Control	⚠️ Basic	✅✅ Configurable (grep, tool outputs, aggressive)
Session Recovery	⚠️ Basic	✅✅ Multiple recovery hooks
Notifications	⚠️ Basic	✅ Background, session, OS notifications
🔒 Security & Permissions
Permission System	⚠️ Via agent configs	✅✅ Fine-grained per-agent
Bash Command Control	⚠️ Basic	✅ Per-command (git: allow, rm: deny)
External Directory Access	⚠️ Basic	✅ Configurable (ask/allow/deny)
Doom Loop Protection	⚠️ Basic	✅ Configurable
🧪 Experimental Features
Experimental Mode	❌	✅ Multiple experimental features
Aggressive Truncation	❌	✅ Experimental
Auto-Resume	❌	✅ Experimental
Dynamic Context Pruning	❌	✅ Experimental (deduplication, supersede writes, purge errors)
📊 Model Support
Model Agnostic	✅✅ Works with any model	✅ Works with any model
Prompt Variants	✅ Per-model prompt files	❌ Same prompt, different models
Multi-Model Orchestration	⚠️ Limited	✅✅ Different models per agent/category
Model Resolution	⚠️ Basic	✅✅ 3-step resolution (override → fallback → default)
Provider Fallback	❌	✅ Automatic provider chain
Opus Dependency	❌ None	⚠️ Recommended for Sisyphus (but configurable)
📖 Documentation & Learning
Documentation Quality	✅✅ Clear, focused	✅✅ Extensive, detailed
Learning Curve	⭐⭐⭐ Moderate	⭐⭐ Steep
Getting Started	✅✅ Simple (context system key)	⚠️ Complex (many agents/features)
Configuration Docs	✅ Good	✅✅ Comprehensive
🎯 Best For
Production Code	✅✅ Code that ships without rework	⚠️ Code that may need refactoring
Team Environments	✅✅ Built for teams	⚠️ Individual-focused
Established Standards	✅✅ Follow YOUR patterns	⚠️ Follow agent's patterns
Token Consciousness	✅✅ MVI principle	⚠️ Requires configuration
Rapid Iteration	⚠️ Approval gates slow down	✅✅ Ultrawork mode
Complex Orchestration	⚠️ Limited parallelization	✅✅ Multi-agent orchestration
Advanced Features	⚠️ Focused, essential	✅✅ Extensive, batteries-included
Customization	✅✅ Behavior (edit prompts)	✅✅ Settings (extensive config)
⭐ Overall Ratings
Token Efficiency	⭐⭐⭐⭐⭐	⭐⭐⭐
Control	⭐⭐⭐⭐⭐	⭐⭐⭐
Repeatability	⭐⭐⭐⭐⭐	⭐⭐⭐
Team-Ready	⭐⭐⭐⭐⭐	⭐⭐
Behavior Customization	⭐⭐⭐⭐⭐	⭐⭐
Settings Configurability	⭐⭐⭐	⭐⭐⭐⭐⭐
Features	⭐⭐⭐⭐	⭐⭐⭐⭐⭐
Parallelization	⭐⭐	⭐⭐⭐⭐⭐
Loop Until Done	⭐⭐	⭐⭐⭐⭐⭐
Advanced Tools	⭐⭐⭐	⭐⭐⭐⭐⭐
Learning Curve	⭐⭐⭐	⭐⭐
Production Ready	⭐⭐⭐⭐⭐	⭐⭐⭐⭐

---

Key Differences Summary

OpenAgentsControl Strengths:

1. ✅ Editable agent prompts - Complete behavior control
2. ✅ Token efficiency - MVI principle, 80% reduction
3. ✅ Approval gates - Human-guided AI
4. ✅ Team-ready - Built for teams with established patterns
5. ✅ Repeatability - Same input → Same output
6. ✅ Context-aware - YOUR patterns loaded automatically

OpenAgentsControl Limitations:

1. ⚠️ Limited parallelization - OpenCode default only
2. ⚠️ Loop until done - Limited (OpenCode default behavior)
3. ⚠️ Advanced features - Focused, not extensive
4. ⚠️ Settings config - Basic compared to Oh My OpenCode

---

Oh My OpenCode Strengths:

1. ✅ Multi-agent orchestration - Full parallelization
2. ✅ Loop until done - Ralph Loop + ulw-loop
3. ✅ Extensive features - LSP, AST-grep, tmux, browser automation
4. ✅ Settings configurability - 15+ options per agent
5. ✅ Background tasks - Configurable concurrency
6. ✅ Advanced tools - Git Master, Comment Checker, Session Tools

Oh My OpenCode Limitations:

1. ⚠️ Baked-in behavior - Cannot edit agent prompts
2. ⚠️ Higher token usage - Multi-agent orchestration overhead
3. ⚠️ Steep learning curve - Many agents, features, options
4. ⚠️ Individual-focused - Not designed for teams
5. ⚠️ No repeatability guarantee - Ultrawork mode is autonomous

---

Decision Matrix

Your Priority	Choose
Edit agent behavior	OpenAgentsControl
Team with established patterns	OpenAgentsControl
Token efficiency	OpenAgentsControl
Approval gates	OpenAgentsControl
Repeatability	OpenAgentsControl
Production code matching YOUR style	OpenAgentsControl
Multi-agent parallelization	Oh My OpenCode
Loop until done	Oh My OpenCode
Advanced features (LSP, AST-grep, tmux)	Oh My OpenCode
Extensive settings config	Oh My OpenCode
"Just do it" mode	Oh My OpenCode
Individual rapid iteration	Oh My OpenCode

---

Key Differences Summary

OpenAgentsControl Strengths:

1. ✅ Editable agent prompts - Complete behavior control
2. ✅ Token efficiency - MVI principle, 80% reduction
3. ✅ Approval gates - Human-guided AI
4. ✅ Team-ready - Built for teams with established patterns
5. ✅ Repeatability - Same input → Same output
6. ✅ Context-aware - YOUR patterns loaded automatically

OpenAgentsControl Limitations:

1. ⚠️ Limited parallelization - OpenCode default only
2. ⚠️ Loop until done - Limited (OpenCode default behavior)
3. ⚠️ Advanced features - Focused, not extensive
4. ⚠️ Settings config - Basic compared to Oh My OpenCode

---

Oh My OpenCode Strengths:

1. ✅ Multi-agent orchestration - Full parallelization
2. ✅ Loop until done - Ralph Loop + ulw-loop
3. ✅ Extensive features - LSP, AST-grep, tmux, browser automation
4. ✅ Settings configurability - 15+ options per agent
5. ✅ Background tasks - Configurable concurrency
6. ✅ Advanced tools - Git Master, Comment Checker, Session Tools

Oh My OpenCode Limitations:

1. ⚠️ Baked-in behavior - Cannot edit agent prompts
2. ⚠️ Higher token usage - Multi-agent orchestration overhead
3. ⚠️ Steep learning curve - Many agents, features, options
4. ⚠️ Individual-focused - Not designed for teams
5. ⚠️ No repeatability guarantee - Ultrawork mode is autonomous

---

The Bottom Line

OpenAgentsControl:

• Philosophy: "Plan first, approve, execute incrementally, ship"
• Best for: Teams with established patterns who need control
• Trade-off: Limited parallelization and advanced features
  Oh My OpenCode:
• Philosophy: "Just do it, agents figure it out, keep going until done"
• Best for: Individuals who want batteries-included orchestration
• Trade-off: Cannot customize agent behavior, higher token usage

---

Both are excellent, but serve fundamentally different needs:

• Need to customize behavior? → OpenAgentsControl
• Need extensive features? → Oh My OpenCode
• Need token efficiency? → OpenAgentsControl
• Need parallelization? → Oh My OpenCode
• Need team-ready? → OpenAgentsControl
• Need loop until done? → Oh My OpenCode