Bring your repository
Start with an existing local Git repository. Create a separate worktree for the task so the main checkout stays intact.
Repository + task branchAn open-source agent workspace for local and cloud developer workflows.
Bring the repository, task, and proposed changes into one place.
Reject invalid email addresses before saving a user profile. Preserve the existing behavior for valid addresses.
Local plan: work in an isolated Git worktree.
This interactive illustration shows the intended experience. It uses sample data and does not execute code or call an AI model.
The real work is keeping context, execution, and review connected.
Coding agents can move quickly. Developers still need to understand what changed, what ran, and whether the result meets the original task. Switching between a chat, a terminal, and a diff makes that harder.
We’re designing rez1 around one complete loop: give an agent a bounded task, follow its work, and make an informed decision about the changes.
Our initial milestone is a local repository-to-diff workflow.
Each step keeps the next decision visible.
Start with an existing local Git repository. Create a separate worktree for the task so the main checkout stays intact.
Repository + task branchDescribe the change, its scope, and how to validate it. Begin with one agent integration and a focused objective.
Context + acceptance criteriaSee the files read, edits proposed, and checks requested. Approve commands before execution and keep actual results visible.
Activity + command approvalsCompare the proposed changes with the task and its test results. Decide what to keep before merging or pushing.
Changes + human reviewThe first runtime is planned for your machine and tools. Isolated cloud environments come after the local loop is dependable.
Explore the architectureA useful review needs more than a patch. The task, tool activity, command output, and test results should be available together.
Read the MVP specificationThe foundation is MIT licensed. Follow the decisions, raise an issue, or contribute to a workflow you want to use yourself.
See how to contributeWe’re developing rez1 workspace, an open-source coding-agent workspace for local and cloud developer workflows. Our first milestone is a local repository-to-diff workflow, with a planned Claude integration for coding tasks and human review of proposed changes.
The integration design connects Claude to a bounded set of repository tools. The local coordinator would handle tool approvals, execution, and the review state.
Read the integration designProvide the objective and relevant repository context.
Read files and propose edits within the task workspace.
Ask to run checks and record their actual outcomes.
Present the diff, checks, and unresolved questions.
The Claude adapter is planned. No integration or partnership is claimed.
Website, source repository, and the specifications for the first workflow.
One complete coding task in an isolated local workspace.
Extend the same task and review model to disposable remote environments.
No agent runtime has shipped yet. We’ll publish release updates in the repository as the implementation develops.
rez1 started with a simple question: what should the workspace around a coding agent look like?
The project began in November 2025. The website and public repository launched in October 2026.
We’re beginning with the foundation rather than promising every workflow at once. The goal is a small, understandable system that helps developers move from a task to a change they can confidently review.
What exists now, what comes next,
and how to get involved.
The current release is the project foundation: this website, the public repository, and design documentation. The local agent runtime is the next milestone. The interactive example on this page is an illustration, not a live agent.
Open an existing local Git repository, create an isolated task worktree, run one agent integration with visible activity and command approvals, and present the resulting diff and checks for human review.
We plan to use Claude for repository exploration and coding tasks through a tool-enabled adapter. The local coordinator will manage tool permissions, command execution, cancellation, and review. This integration is not implemented yet.
The first runtime is local. Cloud execution is a later phase that needs isolated environments, per-task credentials, quotas, and cleanup. The goal is to keep the same task and review experience in both environments.
Yes. The current website and project documentation are available under the MIT license in the public repository. Model services and third-party dependencies have their own terms.
Read the MVP specification, open an issue with a developer workflow you want to improve, or contact the maintainer. Focused feedback and small contributions are welcome while the local workflow takes shape.
Share a workflow, ask about the project,
or help shape the first release.