Watchfire vs Raw agent CLI.

• Agent backend

  Raw agent CLI

  Pinned to whichever CLI you launched.

  Watchfire

  Same CLIs — Claude Code, Codex, opencode, Gemini, Copilot — but pluggable per task or per project.

• Isolation

  Raw agent CLI

  Edits your working tree directly.

  Watchfire

  Per-task git worktree on a watchfire/<n> branch.

• Sandbox

  Raw agent CLI

  Inherits your shell's PATH, ~/.ssh, cloud credentials, and .git/hooks.

  Watchfire

  Seatbelt on macOS, Landlock or Bubblewrap on Linux; sensitive paths blocked by default.

• Parallelism

  Raw agent CLI

  One terminal at a time.

  Watchfire

  One agent per project, many projects concurrently from a single daemon.

• Git workflow

  Raw agent CLI

  Edits the working tree; you commit by hand.

  Watchfire

  Branch + worktree per task; review-then-merge (auto-merge opt-in).

• Autonomy

  Raw agent CLI

  Interactive — whatever the CLI offers.

  Watchfire

  Interactive or autonomous via Wildfire mode.

• Local vs cloud

  Raw agent CLI

  Local.

  Watchfire

  Local — nothing leaves your machine unless you ship it.

• Multi-project

  Raw agent CLI

  Per terminal.

  Watchfire

  First-class.

• Open source

  Raw agent CLI

  Varies (Codex CLI yes; others mixed).

  Watchfire

  Apache-2.0.

• Pricing

  Raw agent CLI

  Free; you pay model API costs.

  Watchfire

  Free; you pay model API costs of the CLI you pick.

When to pick Raw agent CLI.

Pick the raw CLI when you have a single, scoped change to make in a repo you already have open. It is the lowest-friction option — no daemon, no task file, nothing wrapped or filtered between you and whatever the CLI's latest features are. For throwaway repos, exploration, or short conversational work where you want direct access to every flag and surface the CLI exposes, the raw CLI is the right tool. It is also the right tool when you genuinely want the agent to inherit your shell — for example, when you need it to use your real ~/.ssh keys to push, or your cloud credentials to deploy.

Can they coexist?

Yes — trivially. Watchfire calls the same CLI binaries you already have installed. Keep using claude, codex, or opencode directly in any terminal; Watchfire just adds a layer that wraps them in a worktree and sandbox when you want that. The two never fight: the raw CLI works on your working tree, Watchfire works in its own worktree under .watchfire/worktrees/. A practical pattern is to use the raw CLI for the edit you're doing right now and Watchfire for the task queue you want to come back to later.

Side by side.

The same intent expressed in each tool’s idiom.

# In your repo, in your shell
cd ~/code/my-app
claude "Fix the 500 on empty items[]"

# Same agent, same model, behind safety rails
watchfire task add "Fix 500 on empty items[]"
watchfire run all

Related reading.

• Worktree isolationWhy each task runs on its own branch.
• SandboxingSeatbelt and Landlock — what's blocked, what isn't.
• Migrating from raw CLI to WatchfireA long-form walk-through of the first hour.
• See the full comparison matrixEvery adjacent tool, side by side.
• Start with a ready taskCopy-paste starter task YAMLs for the most common work.