OpenCode vs Codex CLI: 161K vs 83K Stars, GPT-5.5 vs Scout Agent (2026)

// OpenCode (anomalyco/opencode) uses Vercel AI SDK
// Client/Server architecture with Hono HTTP API
// v1.15.0 (May 15, 2026), 161K stars, 910 contributors

// Key architectural components:
├── packages/
│   ├── core/           // Agent orchestration (TypeScript)
│   ├── tui/            // Bubble Tea terminal UI
│   ├── web/            // Desktop app (Tauri: macOS, Windows, Linux)
│   └── opencode-ai/    // Model abstraction layer
├── Model Layer: Vercel AI SDK (75+ providers)
├── HTTP API: Hono (enables mobile/remote control)
├── Events: Effect-based event system (v1.15.0)
├── Database: SQLite
├── Desktop: Tauri app with session management
├── New: DigitalOcean OAuth, NVIDIA billing, .well-known/opencode
└── Pricing: Free BYOK | Go $10/mo | Black $200/mo

// Agent modes:
// - "build": Full tool access (default)
// - "plan": Read-only, asks permission
// - "general": Subagent for complex tasks
// - "explore": Read-only codebase navigation
// - "scout": Repo research, docs lookup (v1.14.42)

// Codex CLI (openai/codex): Monolithic Rust implementation
// v0.130.0 (May 8, 2026), 83K stars, 449 contributors

// Key architectural components:
├── codex-rs/
│   ├── core/           // Agent logic + prompts
│   ├── cli/            // Terminal interface + Vim mode (v0.129)
│   ├── mcp-server/     // MCP tool server + plugin marketplace
│   └── exec/           // Command execution + sandbox hardening
├── Models: GPT-5.5 (recommended), GPT-5.4, GPT-5.3-Codex-Spark
├── Cloud: Sandboxed task execution (per-task isolation)
├── Browser: Chrome extension for live browser sessions
├── Remote: codex remote-control for headless operation
├── Mobile: ChatGPT app connection to desktop Codex
├── Hooks: Pre/post-compaction hooks, /hooks browser
├── Goals: Persisted /goal workflows for multi-day tasks
├── Pricing: Plus $20/mo | Pro $100/mo | Pro $200/mo
└── Auth: Amazon Bedrock support, access tokens for CI

// Execution modes:
// - Local: Runs on user machine (Vim mode, voice input)
// - Cloud: Isolated sandbox per task
// - Remote: Headless app-server via codex remote-control
// - Browser: Chrome extension for authenticated web tasks
// - Mobile: ChatGPT app with project/file access

You are a coding agent running in the Codex CLI, a terminal-based
coding assistant. Codex CLI is an open source project led by OpenAI.
You are expected to be precise, safe, and helpful.

## Capabilities
- Receive user prompts and context provided by the harness
- Communicate by streaming thinking & responses
- Make and update plans for multi-step tasks
- Emit function calls to run terminal commands
- Apply patches to modify files
- Use browser via Chrome extension (authenticated sessions)
- Execute persisted /goal workflows across sessions

## Personality
The default personality is concise, direct, and friendly.
It communicates efficiently, keeping the user informed about
ongoing actions without unnecessary detail.

## Model-specific routing (as of May 2026)
// GPT-5.5: Recommended for most tasks (implementation, refactors,
//          debugging, testing, knowledge-work)
// GPT-5.4: Flagship coding model, strong reasoning + tool use
// GPT-5.4-mini: Fast/efficient for subagents and lighter tasks
// GPT-5.3-Codex-Spark: 1,000+ tok/sec real-time iteration (Pro only)

# Example: .opencode/agents/build.md
---
name: build
description: Default agent for development work
mode: primary
model: anthropic/claude-sonnet-4
temperature: 0.7
tools:
  write: true
  edit: true
  bash: true
  webfetch: true
permission:
  edit: allow
  bash: ask
  webfetch: allow
maxSteps: 50
---

You are a coding assistant helping with software development.
Focus on writing clean, maintainable code that follows the
project's existing patterns and conventions.

When making changes:
1. Read relevant files first to understand context
2. Make minimal, focused changes
3. Run tests after modifications
4. Explain your reasoning

// OpenCode edit tool (simplified schema)
{
  "name": "edit",
  "description": "Apply targeted edits to an existing file",
  "parameters": {
    "path": { "type": "string", "description": "File path" },
    "old_string": { "type": "string", "description": "Text to find" },
    "new_string": { "type": "string", "description": "Replacement text" },
    "replace_all": { "type": "boolean", "default": false }
  }
}

// Codex CLI apply_patch tool (simplified schema)
{
  "name": "apply_patch",
  "description": "Apply unified diff patch to modify files",
  "parameters": {
    "patch": { "type": "string", "description": "Unified diff content" }
  }
}

// Key difference: OpenCode uses search-replace semantics
// Codex uses unified diff patches (more git-native)

// Primary Agents (switch with Tab)
├── Build Agent
│   ├── All tools enabled
│   ├── Full file system access
│   └── Default for development work
│
└── Plan Agent
    ├── Read-only by default
    ├── Denies file edits
    └── Asks permission for bash

// Subagents (invoked via Task tool or @ mention)
├── General Subagent (@general)
│   ├── Full tool access (except todo)
│   ├── For complex multi-step tasks
│   └── Runs in parallel child sessions
│
├── Explore Subagent (@explore)
│   ├── Read-only access
│   ├── Fast codebase navigation
│   └── Pattern matching and search
│
├── Scout Subagent (@scout) [NEW v1.14.42]
│   ├── Repo research, docs lookup
│   ├── Dependency-source inspection
│   └── Repository materialization
│
└── Background Subagents [EXPERIMENTAL v1.14.51]
    ├── Tasks continue while you work
    ├── Non-blocking execution
    └── Results available when ready

// Local Execution (default)
User Prompt → Agent Loop → Tool Calls → Local FS
    ↓
   Model: GPT-5.5 (recommended) or GPT-5.3-Codex-Spark (1K+ tok/sec)
   Context: 192K tokens
   Tools: Shell, file ops, MCP, plugin marketplace, hooks
   Input: Text, voice (spacebar), Vim mode (v0.129)
   Memory: Diff-based forgetting + hook-based compaction

// Goals (persisted multi-day workflows, v0.128+)
/goal "Migrate auth to OAuth2" → Stateful across sessions
    ↓
   - Survives restarts and compaction
   - Tracks progress across multiple sessions
   - Useful for migrations, QA passes, content pipelines

// Browser Execution (Chrome extension, May 2026)
Agent → Chrome Extension → Authenticated browser sessions
    ↓
   - Per-site approval controls + allowlist/blocklist
   - Operates across multiple tabs in background
   - Click, navigate, verify visual UI behavior

// Remote / Mobile Execution
codex remote-control → Headless app-server (v0.130)
ChatGPT mobile app → Connected Mac desktop (May 2026)
    ↓
   - Projects, files, credentials accessible remotely
   - Access tokens for CI automation (Enterprise)

// Cloud Execution
User Prompt → Cloud Sandbox → Isolated Container

// opencode.json permission structure
{
  "permission": {
    "edit": "allow",      // allow | ask | deny
    "bash": {
      "*": "ask",         // Default: ask for all
      "git *": "allow",   // Allow git commands
      "npm test": "allow",
      "rm -rf *": "deny"  // Block dangerous patterns
    },
    "webfetch": "allow",
    "task": {
      "general": "allow",
      "explore": "allow",
      "custom-*": "ask"
    },
    "skill": {
      "pr-review": "allow",
      "dangerous-*": "deny"
    }
  }
}

// Permission profiles replaced --full-auto in v0.128
$ codex --profile default     # Ask for destructive ops
$ codex --profile permissive  # Built-in permissive defaults
$ codex --suggest             # Suggest-only, no auto-execute

// Permission profiles support:
- Built-in defaults with sandbox CLI selection
- CWD-aware controls
- Round-trip across TUI sessions and shell escalation
- Active-profile metadata for clients

// Auto-review (v0.128+)
- Eligible approval prompts route through reviewer agent
- Shows risk level before proceeding
- Pre-action review, not post-hoc

// Cloud sandbox (per-task isolation)
- Each cloud task runs in isolated container
- Pre-loaded with repo snapshot
- No access to local filesystem
- Results require explicit approval to merge

// Access tokens (Enterprise, May 2026)
- Non-interactive tokens for CI runners
- Workspace identity for trusted automation

// opencode.json provider configuration
{
  "provider": {
    "default": "anthropic/claude-sonnet-4",
    "fallback": "openai/gpt-5.4",
    "local": "ollama/codellama:34b"
  },
  "agent": {
    "build": {
      "model": "anthropic/claude-sonnet-4"
    },
    "plan": {
      "model": "openai/gpt-5.4"  // Different model for planning
    }
  }
}

// Supported providers (partial list):
// - anthropic/* (Claude models)
// - openai/* (GPT models)
// - google/* (Gemini models)
// - mistral/* (Mistral models)
// - ollama/* (Local models)
// - groq/* (Fast inference)
// - together/* (Open models)
// ... 75+ total providers

// Codex CLI model options (via ChatGPT subscription)
$ codex --model gpt-5.5              # Newest frontier: recommended for most tasks
$ codex --model gpt-5.4              # Flagship: strong reasoning + tool use
$ codex --model gpt-5.4-mini         # Fast/efficient: subagents, lighter tasks
$ codex --model gpt-5.3-codex        # Industry-leading coding for complex SE
$ codex --model gpt-5.3-codex-spark  # Real-time: 1,000+ tok/sec (Pro only)

// Config (~/.codex/config.toml)
[model]
default = "gpt-5.5"            # GPT-5.5 now recommended default
thinking_effort = "high"        # low | medium | high

// Pricing tiers (April 2026 restructure)
// Plus: $20/mo (base Codex access)
// Pro:  $100/mo (5x Plus usage, new April 2026)
// Pro:  $200/mo (20x Plus usage)

// Auth: OpenAI API key, ChatGPT session, or Amazon Bedrock
$ export OPENAI_API_KEY="sk-..."
// Or: AWS console-login credentials for Bedrock (v0.130)