What happens to your code_
Limited beta · read-only GitHub access
Read-only, scoped.
The GitHub app requests read access to code and metadata for the repositories you select. No write access, and no organisation-wide access unless you choose it.
One sandbox per run.
Your repository is cloned only inside an isolated, single-use sandbox, using a short-lived read token minted for that run; the sandbox ends with the run. What persists is the report, the artifacts and the run log — which can include excerpts of files the workflow read.
90 days, or less if you say so.
Reports, logs and artifacts are kept for 90 days per our privacy policy and deleted on request, or when you delete the project.
Commercial APIs. Paid Scans include a Service-improvement license.
During a Scan, model traffic goes through OpenRouter; Midkernel records model/version on the run. Midkernel's policy is not to opt its OpenRouter usage into provider training programmes (that does not control downstream providers). For Scans paid with purchased credit packs, the Terms grant Midkernel a license to use Customer Content and Service data (reports, logs, artifacts, metadata) to improve and develop the Service (routing, ranking, evaluation, related models or systems). That is a rights/purpose grant; Midkernel does not currently operate a dedicated training pipeline. Third parties in the path (GitHub, AWS/compute, OpenRouter, model providers) may retain data under their own policies; Midkernel does not control that.
Runs record what ran.
Each run records its workflow, profile, and the model and version that ran, and the report stays attached to the run's full trace. Pinning the complete input set — repository commit, workflow commit, sandbox image — is being finished across the platform and will be stated here when it is.
Egress limits, being verified.
Sandboxes are designed to reach only what a run needs — the model provider and package registries. We are verifying the enforced egress rules before stating them as a control; until then, treat this as design intent, not a verified guarantee.
Who can see what.
Reports are visible to the members of your workspace. Midkernel staff open a customer's run only for a support request you filed, and the access is recorded.
- compute
- AWS — isolated per-run sandboxes
- hosting
- Vercel
- database
- managed Postgres
- models
- commercial model providers, routed per profile
- Resend
- payments
- native USDC on Base; direct deposits verified on-chain
- analytics
- Vercel Web Analytics (cookieless)
- scheduling
- Cal.com
The legal list of subprocessors lives in the privacy policy and DPA.
Found something in Midkernel?
security@midkernel.com · disclosures are read promptly and fixes are published in the changelog.
No SOC 2 or ISO 27001 report yet, and no independently verified egress or staff-access audit. What we do instead: this page, honest about what is and is not in place, and the trace kept on every run. When a compliance report exists it will be linked here.
Does it need write access?
No. The GitHub app requests read access to code and metadata only. Continuous runs, when they ship, will comment on pull requests through a separate, optional permission.
Can I run it on a private repository?
Yes; that is the normal case. The sandbox sees the repository only for the duration of the run.
Can I run Midkernel in my own cloud?
No hosted private-cloud option. A self-hosted runner is how the platform is designed to run anywhere you like; that repository is not public yet.