Start inside your website
Onboarding discovers your framework and content folders. Choose writable roots, your agent, validation commands, and delivery policy.
Free & open source / BYO coding agent
Your coding agent.
Your website.
Continuous SEO improvements.
RefreshAgent uses your installed Codex or Claude Code to find SEO issues, make scoped changes, run your checks, and leave the results ready for review. Run it locally or schedule unattended work.
macOS & Linux · No Rust compiler · No RefreshAgent account required
- 01Inspect the evidenceVerify the finding against your content and templates.
- 02Make a scoped changeYour coding agent works in an isolated Git worktree.
- 03Run your validationUse the build or checks you configured during setup.
- 04Leave a reviewable diffReview is the default. Inspect the branch and logs.
Workflow overview · execution uses your own agent plan
Local executionIsolated Git worktreesYour validation commandsPause & inspectBuilt-in updates
From repository to review
Give routine SEO work
a place in your workflow.
Built for developers and technical website owners whose content lives in Git, including Markdown and MDX sites.
Work from inspectable findings
The local scanner flags missing titles or descriptions, duplicate titles, and empty links. Your agent verifies the template before editing.
Schedule on your machine
Install a user-level systemd timer or macOS LaunchAgent explicitly. Due checks run one task at a time. Sleeping machines resume checks when available.
Actual CLI output
See a local scan
become a task preview.
Recorded with RefreshAgent against a small example Git repository. This recording shows a scan and dry run; no coding agent ran and no files were edited.
Select Play recording to replay the captured command output.
Recorded local output · terminal commands and results · no simulated traffic gains
Read the recording and reproduce it
The fixture and capture script are in the website repository. The next step on your own site is an authenticated agent run, followed by inspection of its validation logs and diff.
Autonomy you can inspect
Set the boundaries.
Keep the decision.
Review comes first
Validated changes stay on an isolated branch by default. Optional commit or pull-request delivery is explicit. Nothing is automatically merged or published to your CMS.
Validation gates delivery
Configure at least one check before editing. Failures, interrupted runs, and work awaiting review pause scheduled tasks. Inspect history, logs, and diffs before resuming.
Use trusted repositories and appropriately configured coding agents. Your agent and build commands retain their own permissions; the worker is not a security sandbox.
Start locally / no signup
One command.
Your first repository.
The installer downloads a matching release, checks its SHA-256 and version, and installs into your user directory without sudo.
macOS & Linux · Intel / Apple Silicon / ARM64
curl -fsSL https://raw.githubusercontent.com/RefreshAgent/refreshagent/main/install.sh | shHomebrew
brew tap RefreshAgent/tap https://github.com/RefreshAgent/refreshagent brew install RefreshAgent/tap/refreshagent
- Open a new terminal after installation.
- Change into your website's Git repository and run
refreshagent. - Choose your content roots, agent, validation command, and review policy.
Git is required for repository work. Authenticate Codex or Claude Code to execute improvements; local scans need neither. Pull-request delivery also needs GitHub CLI. Model usage follows your agent's own plan.
Linux x86_64 releases require Ubuntu 22.04-era glibc or newer; Linux ARM64 requires Ubuntu 24.04-era glibc or newer. Standalone binaries check for updates daily before work starts. Homebrew uses brew upgrade refreshagent.
Optional RefreshAgent Cloud
Give your agent
Search Console evidence.
Enrich local findings with clicks, impressions, and search positions. Performance evidence helps prioritize the issues already found in your repository.
Connect Google for Cloud data ↗Cloud account and data allowance apply. See current plans.
Useful on its own. Better with evidence.
- Free local worker
- Scanning, BYO-agent execution, validation, history, Git delivery, and local scheduling.
- Optional managed data
- Search Console metrics mapped to your content files. No source files or prompts are submitted by the worker's Cloud sync.
Today's setup uses an account API key, explicit URL-to-file mappings, and refreshagent cloud sync. Sync is manual; there is no automatic Cloud refresh in the scheduled service.
Before you install
What websites does it support today?
Execution currently supports content stored in a Git repository. The scanner focuses on tracked Markdown/MDX and empty Markdown/HTML links. API and mixed content can be recorded during onboarding, but CMS/API execution adapters are not implemented.
Can it work without a RefreshAgent account?
Yes. Local onboarding, scans, agent execution, validation, history, delivery, and scheduling are standalone. Only the optional managed Cloud data connection needs an account.
Whose AI tokens does it use?
Your installed Codex or Claude Code handles authentication and model usage under its own plan. Connecting Cloud data does not switch the local worker to RefreshAgent-paid reasoning.
How do I schedule or pause it?
refreshagent service install --preview refreshagent service install refreshagent pause refreshagent resume
Installation is explicit. The service runs on your machine, using your existing agent setup. Review pending work before resuming; paused schedules do not cancel an already active run.
Does the open-source worker collect telemetry?
No telemetry is sent by the standalone worker. Optional Cloud calls use that service's account and quota policies. This website measures navigation and button interactions separately.
Will it guarantee more traffic?
No. Findings are candidates to verify, and priority scores are transparent heuristics. Review whether a change helps your content and evaluate outcomes using actual performance data.