Providers: Claude, Codex, OpenCode

Tide Commander supports three AI coding CLI providers. Each agent you spawn is backed by exactly one provider. You can mix providers on the same battlefield — a Claude agent and a Codex agent working in parallel is a normal setup.

Provider overview

Claude Code

Default and most capable. Full model selection (Opus, Sonnet, Haiku), configurable reasoning effort, CLAUDE.md support, session resume via --resume. Best for complex, long-running tasks.

OpenAI Codex

OpenAI’s coding agent CLI. Sandbox isolation, workspace-write mode, and --full-auto support. Uses GPT-based models. Good for code generation tasks in isolated environments.

OpenCode

Open-source coding agent. Reads .claude/CLAUDE.md natively. Session resume via -s <sessionId>. A lightweight alternative when you don’t need Claude or Codex.

---

Claude Code

Setup

Install Claude Code from claude.ai/code, then authenticate:

claude login

Verify the CLI is in your PATH:

claude --version

Models

Model	Label	Context	Best for
opus[1m]	Opus [1M]	1M tokens	Very long tasks, large codebases
claude-opus-4-7	Opus 4.7	200K tokens	Complex reasoning, flagship quality
sonnet	Sonnet	200K tokens	Balanced speed and quality
haiku	Haiku	200K tokens	Fast, lightweight tasks

The default selection in the Spawn dialog is Opus [1M].

Effort levels

Effort controls how much reasoning Claude applies before responding:

Level	Description
Low	Minimal reasoning, fastest output
Medium	Balanced
High	Deep reasoning (CLI default)
X-High	Extra reasoning — Tide Commander default
Max	Extended thinking, longest latency

The default selection in the Spawn dialog is X-High.

Session resume

Claude sessions are stored in ~/.claude/projects/. Tide Commander reads these JSONL files to display conversation history and resumes the session with --resume <sessionId> when you restart an agent.

---

OpenAI Codex

Setup

Install the Codex CLI and set your API key:

npm i -g @openai/codex
export OPENAI_API_KEY=sk-...

Verify:

codex --version

Spawn options

When you select Codex as the provider in the Spawn dialog, you get Codex-specific options:

• Full Auto — Runs without pausing for approvals (--full-auto)
• Sandbox mode — Workspace-write isolation (default) or network access
• Approval mode — on-request (prompt when needed) vs never
• Search — Enable web search tool

The default Codex model is gpt-5.3-codex.

Codex sandboxing

Codex runs in a workspace-write sandbox by default. It can read and write files within your working directory but cannot make network requests unless you enable the search option or change the sandbox mode.

---

OpenCode

Setup

Install OpenCode:

npm i -g opencode-ai

Or follow the opencode install guide. OpenCode reads .claude/CLAUDE.md natively, so any existing Claude configuration in your project applies automatically.

Models

OpenCode supports multiple model providers. Set the model string in the Spawn dialog (e.g., minimax/MiniMax-M1-80k). The default selection is minimax/MiniMax-M1-80k.

Any model string accepted by the OpenCode CLI can be entered here.

---

Mixing providers

You can run agents from different providers simultaneously. A common pattern:

• A Claude Opus [1M] boss agent that has full context on the project
• A Codex worker agent in a sandboxed environment for safe code generation
• A Claude Sonnet agent for quick file inspection tasks

The Boss can delegate to any subordinate regardless of its provider.

Provider vs. model

The provider determines which CLI binary runs (claude, codex, opencode). The model is a flag passed to that CLI. You change the provider once at spawn time; you can change the model later from the agent edit panel.