Security at AcruxCore.

Data isolation

Every record in AcruxCore is scoped to a team. The platform enforces this on a shared database with row-level access controls, so one team can never read or write another team's prompts, keys, traces, or evaluations. API requests are authenticated per team and authorized on every call.

Roles and access

Every member of a team holds exactly one role — owner, admin, editor or viewer — and the API authorizes each route against that role, rather than relying on the dashboard to hide a button. Reads that cover the whole team, the audit trail included, are limited to owners and admins.

The audit trail

Every write the platform performs is recorded with the person behind it: prompts, versions and aliases, tools and their bindings, members and invites, API keys, gateway credentials, virtual keys, budgets, secrets, and the team's own trace settings. An owner or admin can read the whole team's trail and filter it by area, by a single event, or by the person who did it — including someone who has since left the team, since their events stay in the record. Secrets appear by name only; a secret's value is never written to the trail.

Reading the team-wide trail needs a signed-in session, not an API key: a record of what people did should not be readable by a program holding a key. Two things it does not do yet — there is no CSV or JSON export, and no retention window to configure, so every event is kept. See Read the team audit trail for what each area records.

Your provider keys

You bring your own model-provider keys. Provider credentials and other secrets are encrypted at rest and are never returned in full through the API or the dashboard once saved. The gateway uses them server-side to route your calls and never exposes them to the browser.

Encryption

• In transit: all traffic to the API, gateway, and dashboard is served over TLS.
• At rest: secrets and provider credentials are encrypted before they are stored.

Payload capture is under your control

Traces record model, tokens, latency, and cost by default. Capturing full request and response payloads is a per-team setting you own — turn it off entirely, or scope it per call — so sensitive prompt and completion content is only stored when you choose.

Self-hosting

When data residency or isolation requirements go beyond the hosted platform, you can run the entire stack yourself — the API, gateway, and dashboard — against your own database and your own provider keys. Nothing leaves your infrastructure.

Reporting a vulnerability

We welcome responsible disclosure. If you believe you have found a security issue, email [email protected] with steps to reproduce and any relevant details. Please give us a reasonable window to investigate and remediate before any public disclosure, and avoid accessing or modifying data that is not your own while testing. That is our single public inbox — security reports are triaged ahead of everything else that arrives there.

We will acknowledge your report, keep you updated on our progress, and credit you once the issue is resolved if you would like.