---
title: "Security: what runs where | Skene"
description: "Skene OSS runs on your machine and reads your codebase read-only. Skene Cloud is the managed version: four permissions, and what each one is for."
url: https://www.skene.ai/product/security
doc_version: "a70667e47978"
last_updated: 2026-08-29
---

Security

# What runs where.

This is the page to forward to your engineering lead or your security reviewer. What Skene can reach depends on which of the two products you run, so each one is answered separately below.

[Start free](https://www.skene.ai/signup) [Contact us about your security review](https://www.skene.ai/contact)

Skene OSS requests no access. Skene Cloud is scoped per repository.

- **Skene OSS runs entirely on your machine**
  MIT-licensed, no Skene account, reads your codebase read-only.

- **Never pushes to your default branch**
  Changes arrive as pull requests you review and merge yourself. No personal access tokens.

- **Supabase starts read-only**
  Schema introspection only. No row data, no table counts, never writes.

Supabase PostHog Mixpanel BigQuery Resend

Access scopes / GitHub App 4 permissions

- **Contents**: Read & write

- **Pull requests**: Read & write

- **Issues**: Read & write

- **Metadata**: Read

Skene OSS

### Runs entirely on your machine.

Reads your codebase read-only, writes skene-context/journey.yaml into your repository. With lmstudio or ollama, nothing leaves the machine.

Skene Cloud

### The managed version, plus what OSS does not do.

Adds pull request review, the journey canvas, and the record of launch checks. The MCP server is hosted; your coding agent connects with a workspace API key.

Access scopes

## Four permissions, and what each is for.

Skene OSS requests no access. Skene Cloud connects to GitHub through the Skene GitHub App, scoped to the repositories you select.

- Contents
  Read & write. Reading your code and `skene-context/` files; committing updates back through pull requests.

- Pull requests
  Read & write. Reading diffs, posting reviews, opening fix pull requests.

- Issues
  Read & write. Responding to `/skene fix` comments.

- Metadata
  Read. Listing the repositories the installation can access.

## Skene Cloud makes two connections.

A workspace owner adds both. Supabase takes a production and an optional development project, assigned separately. GitHub holds one repository and branch.

- Exactly two roles, owner and member. No permission matrix to configure.

- Deploys and every write target the production project only.

**Integrations**/skene-dashboard

2 connected

- Supabase
  connected

- Production acme-production
  assigned

- Development database
  not assigned

- GitHub acme/webapp main
  connected

Data retention

## Skene OSS keeps nothing.

Skene Cloud purges the interaction events Flows captures after 90 days; a workspace owner can delete them sooner. Detailed personal-data retention times are available on request.

Settings / Data retention

| What | Kept for |
| --- | --- |
| Skene OSS output | Nothing |
| Skene Cloud personal data | Account duration |
| Flows interaction events | 90 days |
| Everything, on request | Immediately |

Third parties in the path

## What each one does, and what it sees.

The third parties below are in the path for Skene Cloud.

| Service | What it does | What it sees |
| --- | --- | --- |
| Vercel | Application hosting for the dashboard and the API | Request traffic, session cookies, API payloads |
| Supabase | Authentication, Postgres and storage | Accounts, workspaces, artifacts, encrypted secrets |
| OpenRouter | Gateway for the managed LLM analysis | Prompts, which can include repository and schema context |
| OpenAI | The default managed model, `openai/gpt-5.6-terra`, reached through OpenRouter | The same analysis prompts |
| Google | Gap-analysis models and Skene Agent, through OpenRouter, Gemini and Cloud Run | Analysis prompts and agent chat context |
| Resend | Transactional email from support@skene.ai: invites, welcome, onboarding, reminders and quota warnings | Email address and message content |
| Stripe | Billing and payments | Customer and payment metadata. Card details go to Stripe, not Skene |
| GitHub | The Skene GitHub App: link, webhooks, PR reviews | Repository metadata, PR diffs, review comments |
| Upstash Redis | Buffering on the event ingest path, when it is configured. Without it events buffer in Postgres instead | Buffered analytics events awaiting their scheduled drain |
| Upstash QStash | Fanning the scheduled ingest job out per workspace, when it is configured. Without it the job runs inline | Workspace identifiers, and the scheduling of the work rather than its contents |

Certification

## Not SOC 2 certified today.

SOC 2 certification ROADMAP

It is on the roadmap and not yet started, and there is no date to give you. If a certification is a hard requirement for your procurement, say so early. [Contact us](https://www.skene.ai/contact) and we will answer your security review directly.

## What you keep if we go away.

- Your product data stays in the database you already own.

- Uninstalling the GitHub App withdraws all four permissions at once, from GitHub rather than from us.

- Skene OSS is MIT-licensed and public: if Skene Cloud stopped tomorrow, the tool still runs.

[Start free](https://www.skene.ai/signup) [Contact us about your security review](https://www.skene.ai/contact)

## Sitemap

- [Markdown sitemap](https://www.skene.ai/sitemap.md): every page on this site, with its markdown counterpart.
- [Index for language models](https://www.skene.ai/llms.txt): the same set as a link-dense index.
- [Full corpus](https://www.skene.ai/llms-full.txt): reference content concatenated as plain text.
