Trust
What RidgeLine sees. What it never takes.
RidgeLine is designed around one boundary: usage metadata comes in, and your code stays where it is. This page states the boundary exactly, because your engineers will read the telemetry change before they approve it, and they should.
- Token counts for each agent session: input, output and cached, the measured figures the ledger is built on.
- Session and model identifiers, so usage can be attributed to the agent and configuration that produced it.
- Change metadata from a read-only GitHub App: commit hashes, pull request titles, numbers and descriptions, file paths, branch names, timestamps, author identity as recorded by git.
- Harness configuration, read in place. To grade a repository's guardrails we inspect steering files, checks and review gates read-only, inside your estate. The grade leaves; the files do not.
- Timestamps and repository names, which join usage to the work it became.
- The country a session's request came from. Taken from the network layer, this answers which jurisdictions your code has been worked on from. It is the request's origin, disclosed honestly as that and nothing more; it is not, and cannot be, a model's own hosting location.
- A copy of your code. Nothing is copied out of your estate, nothing is stored on our side, nothing is used for training, and nothing is passed to any other system.
- Prompts or completions. The telemetry carries usage figures only, not the text sent to or received from an agent.
- Secrets or credentials. Neither integration is designed to receive keys, tokens or environment variables.
- Keystrokes, screens or behaviour. RidgeLine meters work, not people. There is no ranking of individuals.
Boring, in the ways that matter.
UK-resident data
Primary infrastructure, including the database that stores your data, runs in AWS eu-west-2 (London). Sub-processors and transfer safeguards are listed in the Data Processing Addendum, available on request.
Pseudonymised mode
Telemetry can send a stable pseudonymous identifier instead of a name and email. Attribution still works; identity stays inside your organisation unless you choose to map it back.
Not for performance management
RidgeLine rates repositories and configurations, never people. The person-level view exists to route support, not scores, and the product is designed to satisfy a works-council review.
Tenancy isolation
Every read and write is scoped to your organisation at the query layer. Data is encrypted in transit and at rest, and secrets are held in a managed store, never in source control.
Governs systems, not people
Governance, in RidgeLine, means assurance that AI-assisted software is built under known controls, at accountable cost, by accountable people. Every control, a pull request check, a policy binding, a jurisdiction rule, acts on the harness and the work it produced. None of it inspects a person.
What we do not yet have.
We would rather tell you plainly what is in progress than imply we already have it. Cyber Essentials certification is planned as the first step on our trust roadmap. ISO 27001 follows when deal size justifies it. We do not claim SOC 2 before we have it. The legal framing of our trust pages is drafted and awaiting solicitor review.
Our Terms of Service, Privacy policy and Data Processing Addendum are drafted and in review; until they publish here, we share them on request at hello@ridgeline.cloud.
Security concerns reach us at security@ridgeline.cloud. We acknowledge good-faith reports and credit researchers who report responsibly, with their permission.
Start with a read-only fortnight.
The two-week audit is read-only from the first minute to the last. Your engineers approve every change that enables it.
Two weeks. Read-only. One number.