Your agent writes it. You get the last edit.
Agents write commit messages, PR text and emails in a style readers now recognise: padded, formal, summed up at the end. reright holds each text until you have cut it down to how you would write it. The agent then sends exactly that, or stops.
Works with your AI agent, whichever model or tool, in the cloud or on your machine.
Free plan: 100 messages a month, no card. Hooks that block unapproved sends are tested in Claude Code. Codex CLI, Gemini CLI, Copilot and Cursor are built from their documentation and not yet tested in the real tools. Other clients are told to submit first, and nothing forces them.
-
Commit message
Agent draft, with your cuts and additions
feat(sync): implement a robust retry mechanism to enhance reliability- Add exponential backoff to the retry loop - Improve error handling for rate limit responses - Ensure the invoice sync is more resilient This change enhances the overall reliability of the sync.You approve
fix invoice sync retry
-
Pull request description
Agent draft, with your cuts and additions
EnhanceImprove login redirect handlingfor improved user experience## Summary This PR fixes an issue where users who followed a deep link were redirected to the home page after logging in.It ensures a seamless authentication experience.## Changes -UtilizeUse the `next` parameter to return users to their original page -Added tests to ensure robust coverageAdd testsOverall, this change improves the reliability of the login flow.You approve
Improve login redirect handling ## Summary This PR fixes an issue where users who followed a deep link were redirected to the home page after logging in. ## Changes - Use the `next` parameter to return users to their original page - Add tests
-
Client email
Agent draft, with your cuts and additions
Update regarding yourYour invoice export requestDear Marta Silva,Hi Marta,I hope this email finds you well.I am writing to inform you that the invoice export you requested has been completed.The invoice export you requested is done. The file is available in the shared folder.Please do not hesitate to contact me should you have any questions or require further assistance.Let me know if you have any questions. Bestregards,You approve
Your invoice export request Hi Marta, The invoice export you requested is done. The file is available in the shared folder. Let me know if you have any questions. Best,
-
Customer support reply
Assistant draft, with your cuts and additions
DearHi Mr. Costa,thank you for contacting us regarding your recent order.thanks for getting in touch about your order.We sincerely apologize for the inconvenience caused by the damaged item. We would like to assure you that we take such matters very seriously.Sorry your item was damaged. We haveinitiatedstarted a replacement, which will bedispatchedsent within 2 business days.Please do not hesitate to reach out should you require any further assistance.Reply here if you need anything else.You approve
Hi Mr. Costa, thanks for getting in touch about your order. Sorry your item was damaged. We have started a replacement, which will be sent within 2 business days. Reply here if you need anything else.
How it works
The loop you use every day.
-
The agent writes, then waits
It submits its text with a note on who will read it. Nothing is sent until you have reviewed it.
-
You edit it into your words
You get a notification, read the context next to the text and cut what you would never write.
-
Approve, and the next one opens
Press Ctrl+Enter. The agent sends exactly what you approved, or stops if you reject. The next draft is already open, with the cursor in the editor.
Get started
Five steps, and setup takes a couple of minutes.
-
Sign up
Create an account on the free plan. No card.
-
Copy the setup block
The dashboard shows it, with a one-time code.
-
Paste it into your agent
Claude Code, Codex, Gemini CLI, Copilot or Cursor. The installer finds the agents on your machine and sets up hooks, the MCP connection and the rules text.
-
Approve the test draft
A sample draft appears in reright. Press Ctrl+Enter.
-
Done
Your agent now submits its text before it sends. The docs table shows which agents are blocked by hooks and which are only told to ask.
You paste the block from the dashboard: > Set up reright: fetch https://app.reright.it/setup.md and follow it. Setup code: XXXX-XXXX The installer prints a summary: Summary claude configured codex configured git installed Test draft sent. Approve it in reright.
reright uninstall puts them back.Works where you already work
reright is an MCP server. Any AI client that can call MCP tools can use it, from any model provider, in the cloud or running locally. Clients that support MCP include Claude Code, Codex, Cursor, Gemini CLI, VS Code with Copilot, Claude Desktop, Claude.ai and ChatGPT.
Clients
Coding agents. Editors with an agent mode. Chat assistants, through their custom connector settings. Any other client that can call MCP tools.
Text
Commit messages, PR text, client emails, support replies, posts, reports and notes. Anything a person will read under your name.
One install sets up hooks that block a send until you approve the text, in Claude Code, Codex CLI, Gemini CLI, GitHub Copilot and Cursor. Claude Code is tested end to end. The other four are built from their documentation and checked only against documented payloads, not against the real tools, and we do not yet know whether their hooks still run in auto-approve modes. A git commit check covers commit text for any agent that runs git. Other clients get the MCP tools and a rules text, and nothing stops a client that ignores it. The docs table shows each one.
Why not approve where the agent runs?
You can approve in the agent's own chat or terminal, or in a remote-control app. reright is a separate place to review text, and it differs in these ways.
| Topic | In the agent's own chat or terminal, or a remote-control app | In reright |
|---|---|---|
| Context | The text sits in a transcript, among everything else the agent printed. | The text is shown next to its context in a fixed structure: what it is, why, who reads it and what changed. |
| One inbox | Each agent asks in its own window, and you have to be watching that window. | Drafts from several agents land in one queue with their context, grouped by agent and run, with filters, whichever tool or vendor they came from. |
| Editing | You copy the text out, edit it and paste it back, or ask the agent to redo it and wait. | You edit the suggested text directly and press Ctrl+Enter. Reviewing a queue works from the keyboard. |
| Alerts and access | You find out when you look, and usually from the machine or session the agent runs in. | Desktop and phone alerts when something is waiting. It runs in the cloud, so you can review from any device. |
| What is sent | Your yes is not tied to the exact text that gets sent. | What you approve is what goes out. Where hooks run, the send is blocked if one character differs and a reject stops the agent. Hooks are tested in Claude Code and built but untested in four more agents, and a git check covers commits from any agent. Other clients only tell the agent to do this. |
| History | The proposal and your edits scroll away with the session. | You can see what the agent proposed, what you changed and what you rejected, for as long as your plan keeps text. |
Pricing
Start free. Move up when you pass 100 messages a month.
-
Free
$0
100 messages a month
Text kept 1 day, fixed
Start free -
Starter
$5 a month
1,000 messages a month
Text kept 1 to 90 days, default 30
Choose Starter -
Pro
$25 a month
10,000 messages a month
Text kept 1 to 90 days, default 30
Choose Pro -
Unlimited
$100 a month
Fair use, up to 500,000 a month
Text kept 1 to 90 days, default 30
Choose Unlimited
- A message is one draft submitted for review. Reviewing drafts you already have never counts.
- Months are calendar months in UTC. Fair use is defined in the terms.
- Prices are in US dollars. VAT and other taxes are added at checkout where they apply.
- Upgrades take effect at once and downgrades at the end of the billing period. Moving down to Free deletes text older than one day, and the dashboard asks you to confirm first.
Your text and how long we keep it
Text is deleted from the live database at the retention you choose, and from every backup within 48 hours after that.
Free keeps text for 1 day. Starter, Pro and Unlimited let you choose 1 to 90 days, default 30. Drafts that nobody decides expire on the same schedule. Counts and billing records stay.
Text is encrypted at rest and in transit. Stored text is encrypted with AES-256-GCM under a separate key for each user, and a master key kept outside the database wraps those keys. Your email address, plan and timestamps are not encrypted in the database.
This is not end-to-end encryption. The server decrypts your text to show it to you and to check approvals, and interpt runs the server. The encryption protects text in a stolen database or backup, not against the operator.
Backups reach back about a day. We can restore the service after a failure, but not text that was deleted more than a day or so ago. The privacy policy has the details.
Questions
Short answers on enforcement, data and billing.
Does reright detect or rewrite AI text?
No. It does not analyse text or change it. The agent's draft goes to you, you edit it yourself, and only the text you approve is sent.
Can my agent get around it?
Yes, in several ways. Where hooks run, they block git commits, GitHub CLI text, Gmail sends and text typed into the browser unless you approved that exact text in the last 24 hours. But the hooks run on your machine, so anyone who can edit the hook files or set RERIGHT_BYPASS can switch them off. git commit --no-verify can get past the git check in some cases, and commands such as git commit-tree skip it. Auto-approve modes may skip hooks, and we have not checked that for the four untested agents. Chat assistants and other clients can ignore the rules text. The docs list what bypasses it.
Does it work with agents other than Claude Code?
Yes. The install finds Claude Code, Codex CLI, Gemini CLI, GitHub Copilot (CLI and VS Code) and Cursor on your machine and sets up each one with hooks that block unapproved sends, the MCP server and a rules text. Claude Code is tested end to end. The other four are built from their documentation and not yet tested against the real tools. Any other client that can call MCP tools gets the tools and the rules text, and nothing forces it to use them.
What counts as a message?
Each call to submit_for_review counts once, in the month it was made. Over quota, new submissions fail with a link to upgrade and the date your quota resets. You can still review, approve and reject drafts that already exist.
Who can read my drafts?
You can. The server decrypts them for you and for the approval check. Nobody at interpt reads them as part of running the service, but an operator with full access to the server could, which is why we do not call it end-to-end encryption.
Does the site track me?
No. This site sets no cookies and loads no third-party scripts, analytics or fonts. Your theme choice is saved in your browser and never sent anywhere. The app at app.reright.it uses a session cookie so you stay logged in.
Who do I pay, and who sends the invoice?
interpt sells reright directly and issues the invoice. Stripe takes the card payment and calculates the tax at checkout. Card details go to Stripe and never reach us.
How do I cancel?
Downgrade to Free from your dashboard. It takes effect at the end of the period you paid for. Text older than one day is then deleted at the next hourly purge and cannot be recovered. The dashboard shows how many drafts that affects and asks you to confirm first. Deleting your account removes your data and cancels the paid plan.
Try it on your next commit message
Setup takes a couple of minutes.