Security & Privacy
Your runner. Your keys. No service in between.
Secure by Architecture
The safest service is the one that doesn't exist
Runs on Your Runner
The Action runs inside your own GitHub Actions job. PR and issue content goes from GitHub to your LLM provider and into the outputs you choose. No ReleaseRay service ever sees it.
Secrets Stay in GitHub
Your LLM key lives in GitHub Actions secrets and is sent only to your provider. ReleaseRay stores no credentials anywhere.
Bring Your Own Provider
Use OpenAI, an OpenAI-compatible endpoint you run yourself, or AWS Bedrock in your own account, so inference stays where your policies require.
No Database to Breach
State lives in git: markers in the GitHub Release and CHANGELOG.md record what was published. There is no ReleaseRay database.
Least-Privilege Token
The default GITHUB_TOKEN is enough for most repositories: read access to PRs and issues, and write access only where an output needs it.
Nothing to Rotate on Our Side
ReleaseRay holds no keys of yours. Rotating a key means updating one Actions secret; nothing else has a copy.
Audit Trail in Your Workflow Runs
Every run is recorded in your GitHub Actions history, and the job summary shows the release range, the outputs written, and the tokens used.
Open Source and Reviewable
Every line is public under the MIT license. The example workflows pin the Action to a full commit SHA, so the code you reviewed is the code that runs.
Output Filters Against Prompt Injection
PR and issue text is treated as untrusted. Generated notes are filtered for links, mentions, HTML, and images before anything is published.
Hardened Workflow Examples
The examples avoid expression injection from untrusted input, and a two-job example separates reading from publishing for public repositories that merge outside contributions.
Privacy First
Your data stays in GitHub and with the LLM provider you choose
Nothing Retained Elsewhere
Notes live in your repository and your releases. Files the Action writes exist only for the job unless you upload them yourself.
Delete by Deleting
Remove the notes from a release or CHANGELOG.md, or delete the workflow runs. There is no ReleaseRay account to close.
Internal Notes Stay Internal
The internal persona's notes never go to public surfaces like the GitHub Release or CHANGELOG.md unless you explicitly send them there.
What the Action Reads vs. What It Writes
Read During the Run
Tags and commits from your checkout, and PR and issue titles, bodies, and labels from the GitHub API. They are sent to your LLM provider and nowhere else.
Written Where You Choose
The generated notes, in the GitHub Release, CHANGELOG.md, the job summary, and files in the workspace. Nothing is stored outside your repository and runner.
ReleaseRay does not send your source files to the LLM. It reads PR and issue content and commit metadata, and there is no ReleaseRay infrastructure for any of it to touch.
For Your Security Review
What to tell your security team
No Data Processor
By DesignReleaseRay runs no service, so it never processes or stores your data.
Your LLM Terms
Your ContractInference runs under your own agreement with OpenAI, AWS, or your endpoint.
Open Source
MITThe full source is public and can be reviewed before you adopt it.
Vulnerability Reports
PrivateReported through GitHub's private vulnerability reporting for the repository.
Questions about security?
Read the security policy, or report a vulnerability privately through GitHub