Get a Free SEO Audit or Automation Assessment
CodeWrench · Free AI IDE · Alpha
Your AI Agents. One Free IDE.
Bring Claude Code, Codex, Cursor and your other coding agents into one AI development environment. Run tasks in parallel in separate Git worktrees, see every agent's terminal in one grid, and review code and browser previews before you ship.
Free to use. AI subscriptions, API usage, voice and publishing services are billed separately.
For developers who already use AI coding agents. CodeWrench is in alpha: it is updated often, and some parts are unfinished. Made by Savage Digital Solutions.
No Install. No Sign-In.
Try the CodeWrench IDE.
A browser version of the CodeWrench interface. AI replies and terminal commands are simulated, and nothing is installed. Open a conversation, switch to Code and launch a sample team.
From Task to Reviewed Change
Open Your Project
Choose a repository and connect the coding providers you want to use. Your files, conversations and source control belong to the same workspace.
Give the Agents a Task
Describe the change by text or voice. Start another task in its own worktree when you want work to happen in parallel.
Review and Ship
Read the diff, inspect the browser preview, and ask for changes. Commit the result and open a pull request when it is ready.
What’s Built Today
These capabilities are implemented in the current development build. Published alpha downloads may lag behind it, and some integrations require additional setup.
Your Agents and Models
You stop switching between separate apps, and you keep using the AI subscriptions you already pay for.
Use Claude Code, Codex, Cursor, Grok, OpenCode, Antigravity, Pi and other agents through ACP, a standard agent connection. Choose a provider, account and supported model for the task, and set a permission mode that decides when an agent must ask before it acts. Hand a conversation to another provider with a summary of the earlier context. A Usage page combines token use and estimated API-equivalent cost across your connected providers; the estimates are not your subscription bill. Provider capabilities and setup requirements vary.
Parallel Work and Agent Teams
Agents working at the same time do not overwrite each other's files, because each task has its own copy of the code.
Run independent tasks in branches and Git worktrees (separate copies of the code). Worktrees keep file edits apart; they do not resolve merge conflicts for you. The Code tab supports up to twelve terminal panes per workspace and a team launcher for up to six agents with roles, shared tasks and individual assignments.
Terminals, Tasks and Layouts
Each pane shows whether its agent is Working, Needs You or Idle, so you can see which one needs your attention without checking them all.
Keep real coding CLIs and shell sessions together. Arrange panes, save layouts, broadcast to selected sessions, keep workspace notes, and track cards through To Do, In Progress, Review and Done.
Git and Pull Request Review
You review the work and ship it from the same window the agents work in.
Inspect diffs, commit changes and create or review pull requests. Source control integrations include GitHub, GitLab, Forgejo, Gitea, Bitbucket and Azure DevOps, with supported actions depending on the host.
Browser Previews and Feedback
You check the page an agent built without leaving the app.
Keep a browser beside the conversation. Navigate, inspect elements, take screenshots, record interactions and attach visual feedback while reviewing an interface.
Voice Input and Conversations
You can speak a long brief instead of typing it.
Dictate into a draft and review it before sending. Current development builds also support live agent conversations through ElevenLabs, with an optional OpenAI voice provider. Voice-service usage is billed separately; availability depends on the client and setup.
Skills and Routing Advice
Agents get the right instructions for a task without every skill loading into every session.
Manage skills and plugins, attach relevant skills to a task, inspect reported account allowances and request optional model-selection advice. Review the recommendations and choose your model.
Standing Agents and Scheduled Work
Work continues while you are away, and the results wait for you.
Create agent profiles, standing goals, recurring tasks and check-ins. See next runs and results, pause schedules and review what needs attention. The environment running the work must stay online.
Files and Media Review
You can show an agent exactly what should change.
Attach files and images, preview code and documents, and annotate images or video frames with regions and time ranges. Configured tools can generate images and videos or edit media; external services, local hardware and additional dependencies may be required.
Campaigns and Content Workflows
Agents draft the posts, and you decide which ones are approved.
Organize campaign drafts, captions, platforms and media on an approval board. Export approved posts or connect a publishing service. Publishing is off until configured and explicitly enabled; external-service costs are separate.
Connected Environments
You can check on your agents and steer them from another computer.
Connect another browser or client to a running host through direct pairing, CodeWrench Connect (sign in, no router setup) or desktop-managed SSH. Projects and agents stay on that host, which must remain available. Supported client surfaces differ.
Devices and Desktop Capture
You can watch an agent check its mobile work on a simulator.
Inspect iOS Simulators and Android Emulators beside a thread when the host has the required tools. Optional desktop SnapShots (a shortcut that captures a window) attach the capture to a draft. Device control and desktop capture require setup and permissions.
Get CodeWrench
Installers: Mac (Apple silicon) and Windows 10 or 11. Intel Macs and Linux do not have an installer. Download yours, follow the install steps, then connect your AI providers and a project.
What You Need Before You Start
- At least one coding agent installed and signed in on your computer, such as Claude Code, Codex or Cursor. You can also accept a model someone shared with you.
- Git installed, and a project folder to open. Parallel work in separate worktrees needs a Git repository with at least one commit.
- A CodeWrench account (free). One account links your computers. You create one or sign in after you install.
On first launch macOS and Windows ask you to confirm. The Mac build is not notarized and the Windows build is not code-signed.
CodeWrench has no software subscription fee. Your connected AI, voice and publishing services have their own pricing.
Questions, or something broke? Contact Savage Digital Solutions.
Run Coding Agents in Parallel
A practical Claude Code and Codex workflow with separate Git worktrees, builder and reviewer roles, and a review before merge.
Read the WorkflowCovering CodeWrench?
Get product facts, clean screenshots, a real interface recording and details for a hands-on story or review.
Open the Press Kit