Refresh is an open-source coding agent client built for real software work. It gives you one place to run Claude Code, Codex, and other agent workflows across local repositories, isolated git worktrees, background sandboxes, and a visual desktop UI.
Instead of bouncing between terminal tabs, browser tools, diff viewers, and chat windows, Refresh keeps the full loop in one workspace: plan, execute, inspect changes, and ship.
Why Refresh exists
Coding agents become much more useful when they can work inside a safe, inspectable workflow.
That is the core idea behind Refresh:
- keep code close to the developer
- make every tool action visible
- isolate risky changes from the main branch
- support both local execution and background cloud runs
Refresh is designed for people who want the speed of agents without giving up reviewability, branch safety, or control over their environment.
What makes Refresh different
1. Multiple agents in one product
Refresh is not tied to a single provider. You can run Claude Code, Codex, and other model/provider combinations in the same app and switch between them based on the task.
That matters because planning, editing, browsing, and long-running work often benefit from different models.
2. Local-first execution
Your repository stays on your machine. Refresh is built around local development workflows, so you can point the agent at a real project folder, inspect what it is doing, and keep sensitive code inside your own environment.
3. Git worktree isolation
Each chat can run inside its own isolated worktree instead of touching your main branch directly. This removes a whole class of messy failure modes: accidental edits to the wrong branch, mixed experiments, and hard-to-review changes.
Good tooling removes special cases. Worktree isolation is one of those design choices that makes the normal path safer by default.
4. Visual review before merge
Refresh is not only a chat surface. It includes diff previews, tool execution logs, git activity, file inspection, and rollback-oriented workflows, so you can review what happened before accepting the result.
Agents are faster when their work is easy to audit.
5. Background agents and browser previews
Some tasks do not belong on an active laptop session. Refresh supports background execution in cloud sandboxes, plus live browser previews for dev branches, so work can continue even when your local machine is offline.
Core workflows
Refresh is built around a few practical loops:
- Start with a plan when the task is ambiguous.
- Run an agent in a safe workspace when the direction is clear.
- Watch commands, file edits, and tool calls as they happen.
- Review diffs, git changes, and outputs before shipping.
- Move longer work into background agents or automations when needed.
This is the difference between "AI chat about code" and "an actual environment for coding agents."
Who Refresh is for
Refresh fits teams and individual developers who:
- work across real repositories instead of toy prompts
- want to compare or combine multiple coding agents
- care about branch safety and inspectable changes
- need a desktop-first workflow with optional cloud execution
- want MCP, custom providers, slash commands, memory files, and automation hooks in one place
A simple way to think about it
If a terminal gives you raw power, Refresh gives you operational clarity.
It is a coding-agent workspace that combines:
- the flexibility of local development
- the safety of isolated execution
- the visibility of a visual review layer
- the leverage of modern AI agents
Final thought
Refresh is built around a simple belief: coding agents are most useful when they are visible, controllable, and embedded in real development workflows.
That means fewer magic demos and more practical software shipping.
