A local work surface for coding agents
The work happens
in Rally.
Run Claude Code, Codex, and Antigravity on your Mac. Select the exact parts of an agent response that need work, add your notes, and send one clear instruction back.
The Input Harness
Point to the work.
Then steer it.
Select the exact phrases that need work. Rally keeps every quote beside its comment, preserves their order, and sends the review back without another model rewriting it.
- 01SelectMark the exact output.
- 02AnnotateAdd focused context.
- 03CombineBuild one ordered review.
- 04Send backSet the next direction.
Spaces
Stay oriented as
sessions multiply.
Group related sessions into Spaces. Switch projects and Rally restores the sessions and active thread you left there, while each agent keeps its own provider context.
Your agents · your machine
Keep the tools
you already use.
Rally gives local coding-agent sessions one working surface. It does not replace your provider account, proxy prompts through a Rally model, or merge context between agents.
Run the agents already installed on your Mac.
Start Claude Code, Codex, and Antigravity sessions from Rally and use the accounts you have already configured.
Keep each session’s own context and permissions.
Rally organizes the work around each agent. File access, commands, model behavior, and permission prompts still belong to that provider.
Work Together
Invite review.
Keep the agent in hand.
Share one active session with another Rally app. Guests can watch, comment, or propose a prompt within the role you approve. Nothing reaches the agent unless the host chooses to send it.
Review and decide.
The host runs the agent and sets the boundary for every guest.
- Approve access to one selected session
- Choose Viewer or Commenter permissions
- Ignore, edit, or send proposed prompts
Watch and suggest.
The guest joins through Rally, with only the role and session the host approved.
- Follow the selected session as a Viewer
- Comment on an exact passage as a Commenter
- Propose a prompt for the host to review
A clear sharing boundary
Share one session.
On explicit terms.
The agent stays on the host Mac
Rally launches the supported provider CLI that is already installed and signed in there.
The selected view is encrypted
The shared session view and guest feedback are encrypted before relay. Other sessions and raw files are not transferred.
Roles before actions
The host approves Viewer or Commenter access. Guest feedback never runs automatically.
Provider permissions still apply
Agents can run commands and edit files according to the permission mode set for that provider.
A shared transcript can contain text the agent read from a file. Review the session before sharing and use trusted project folders.
Private beta
Small cohorts,
direct learning.
Rally is opening in small groups while the review loop, provider behavior, and sharing boundaries are still being tested with real work.
Open RallyApple Silicon · macOS 12+ · Signed & notarized
Private Beta
Install Rally
Already have an invitation? Enter your access code to continue.
Invitations for your cohort
Invite another agent user
Save these codes now. Share each with one person who already uses local coding agents. Every code works once.
These codes are shown only here. Rally does not send them to product analytics.
You're in. Your download should start automatically.
Didn't start? Download Rally again →
Installing the beta
- Open the downloaded DMG.
- Drag Rally to Applications.
- Open Rally from Applications and choose Open if macOS shows its standard downloaded-app confirmation.
Rally beta is signed and notarized by Apple, so the unidentified-developer bypass is not required. Future updates install from inside Rally.
Hit a rough edge? Send feedback → It shapes what ships next.
FAQ
Before you install.
What is Rally?
Rally is a native macOS work surface for running local coding agents, reviewing their output, and setting the next direction. It currently supports Claude Code, Codex, and Antigravity.
Who is the private beta for?
People who already use coding agents in real projects and regularly manage several sessions, review long responses, or turn detailed feedback into follow-up prompts.
What do I need to run Rally?
An Apple Silicon Mac with macOS 12 or later, plus Claude Code, Codex, or Antigravity installed and signed in.
Does Rally include model access?
No. Rally uses the provider accounts and command-line agents you have already configured.
Does Rally send my prompts through its own AI model?
No. Local sessions run through the connected agent on your Mac. Rally does not proxy prompts through a Rally-hosted model, and the Input Harness does not use another model to rewrite your review.
Do agents share memory or context?
No. Rally organizes sessions together, but each session keeps its own provider context and permissions.
What is shared in Work Together?
The selected session view and guest feedback are encrypted and relayed. Raw project files, other sessions, and other Spaces are not transferred, though a transcript can contain text an agent read from a file.
Can a guest run my agent?
No. A guest can view, comment, or propose a prompt within the role you approve. The host decides whether anything is sent to the local agent.
Why does the beta require an invitation?
Rally is opening in small cohorts so feedback can stay close to real workflows while agent integrations and sharing boundaries are still being tested.
Will macOS block the beta?
Rally is signed with its Developer ID and notarized by Apple. macOS may show the standard downloaded-app confirmation, but no unidentified-developer bypass should be needed.
Feedback
Shape the beta.
Rally is early. Bugs, rough edges, missing features — tell us and it goes straight to the maker. Email is optional; leave it if you'd like a reply.
Thanks — got it. This is exactly how the beta gets better.
Review the result. Set the next direction.
The work happens
in Rally.
One place to run the session, point to what needs work, and move it forward.