EU‑sovereign CI/CD with Git included.

What you get

Everything in one €30/seat plan. No "Advanced Security" add-on, no Actions metering surprise, no Marketplace trap.

Enterprise access controls

The enterprise controls regulated teams ask for, wired to a real per-org policy bundle. Change any of them, an audit event fires.

• SSO to your IdP. OIDC: Entra ID, ADFS, Okta, Google Workspace, Authentik, Keycloak. Per-org session binding; per-tenant auth-source with tenant-scoped group-team maps.
• Real step-up MFA. acr_values passthrough on the authorize request, plus amr-claim verification on every callback. A signed-in session actually proves MFA happened - not a claim buried at first login.
• SCIM 2.0 provisioning. Push users and groups from Okta, Entra, Authentik, or any compliant IdP. Per-tenant bearer-token auth with dual-token rotation.
• SSH certificate authority. Deploy a fremforge-hosted SSH CA per org, or bring your own. Certificate-anchored keys carry TTL, principals, and revocation independent of the key material.
• Hardware-backed keys. WebAuthn-only for 2FA (no TOTP fallback). Enforceable per-org for regulated tenants.
• IP allowlist. Restrict tenant sign-in and API access to your corporate CIDRs. Enforced at the API gate, not just the UI.
• Email-domain allowlist. Members can only add @acme.com-style addresses matching your patterns. Wildcards supported; OIDC sign-in exempt (the IdP attests).
• Max PAT lifetime. Cap personal-access-token TTL below the 90-day platform default. Per-org override.
• Signed-commits required. Block unsigned pushes at the pre-receive gate. SSH-key signing is Forgejo-native; GPG also supported.
• Org-wide branch protection. One set of defaults for every repository - require a pull request, required approvals and status checks, review from code owners - applied to existing repositories as soon as you save. Weakened repositories are flagged.

Security, in detail

GitHub sells Secret Protection and Code Security as add-ons, $49 a month for each active committer on top of the seat. We include the meaningful part.

• Pre-receive secret scanning. High-confidence patterns rejected before the commit enters immutable history.
• Push protection. Detected secrets refused at push time; hash-based override system with cross-tenant operator review; per-tenant failure-mode toggle (fail-open for dev velocity, fail-closed for regulated).
• Dependency scanning on PRs. Trivy + osv-scanner across the repo's dependencies; findings as PR comments; merge-block at the severity you choose.
• Reachability. Each dependency and image finding is marked reachable, not reachable or unknown, with the file and line or function as evidence, and lists are ranked by it. Go: govulncheck on the binaries in your images. JavaScript and TypeScript: Jelly. Python, Java and Kotlin: import level. Nothing is hidden or auto-dismissed; not blocking on not-reachable findings is your opt-in.
• Known exploited first. Vulnerabilities listed by ENISA's EU Vulnerability Database or CISA's KEV catalogue are badged, ranked above other findings of the same verdict and severity, never given a stretched SLA, and get their fix pull requests first.
• Security fix pull requests. One click opens a pull request that bumps a vulnerable package to the lowest fixing version (npm, Go, PyPI, Maven, Gradle, NuGet, Cargo) or an image's base image; Fix all per repository, or automatically for the findings that matter most. One pull request per package; known exploited first, then reachable, then by severity. They wait for a person's review by default, and merge when checks pass only if you opt in (never across a major version, never during a release freeze). No need to switch on version updates.
• Static analysis (SAST) on PRs. OpenGrep runs on every pull request; findings as PR comment + commit status; on by default.
• License scanning. SPDX-based allowlist per repo; findings on PR; policy inheritable from org default.
• Malware scan on package uploads. ClamAV in front of every package and container image upload, including from CI; rejected uploads never enter object storage.
• SBOM per build. Syft SBOM emitted with every runner-built image, cosign-signed and attached to the release.
• SLSA provenance. Hosted runners emit signed provenance; verify with slsa-verifier.
• Signed commits. SSH-key signing as default (Forgejo verifies natively); GPG also supported. No third-party log fan-out.
• Workflow pinning. One click to rewrite every uses: actions/checkout@v4 in your workflows to the pinned SHA digest.
• Three-level RBAC. Platform-floor controls, org-default policy set by Owner/Admin, per-repo override that can tighten freely.
• Compliance. Compliance report, control evidence and an on-demand signed HTML/JSON vendor attestation of your tenant's audit-chain + security posture. Render from live state, hand to a customer, auditor, or regulator without a screen-share.

Keyless commit signing is available via fremverk's self-hosted Sigstore stack on T Cloud Public eu-de (Fulcio + TSA; no Rekor transparency-log fan-out). Endpoints: sign.frem.sh and tsa.frem.sh. No request transits the public Linux Foundation Sigstore instance. Customers opt in per-tenant at Authentication policy → Require OIDC-signed commits. See docs.frem.sh/get-started/keyless-commit-signing.

Dependency proxy

Public packages and container images come through frem.sh, cached for your organisation and checked against your package policy. On by default for every organisation, and used by the hosted runners.

Known malware (OSV MAL-* advisories) is refused for every organisation, and a package policy in Enforce refuses what breaks it. Cached content keeps being served when an upstream registry is down. If the proxy cannot be reached when a runner job starts, the job installs and pulls from the public registries instead. Docker Hub and Red Hat's authenticated registries use your organisation's own credential, added by an owner. See docs.frem.sh/develop/packages/dependency-proxy.

Ship and operate

The operator surface an SRE team needs to actually run this in production.

• Deployment environments. Named environments (prod / staging / preview) with per-env secrets, per-env branch restrictions, and required-reviewer gates. Approvals visible in the PR checks and the operator console.
• SIEM forwarding. POST fremforge audit events to your Splunk/QRadar/Sentinel/Datadog endpoint. HMAC-signed payload, per-endpoint retention, test-mint from the admin UI, disable per endpoint without re-onboarding.
• Fix-by SLA. Set the days to fix per severity (defaults 7 / 30 / 90 for critical / high / medium). Overdue findings surface as bell alerts and in the security digest.
• Security digest. Weekly (or daily) email digest of open findings + newly-arrived critical CVEs. Per-user opt-in.
• Audit chain. Per-tenant SHA-256 hash-linked chain of every state change. Anchored to WORM storage every 2 minutes, replicated cross-region.
• Audit retention, configurable. Payload retention 90 / 180 / 365 / 730 days per tenant; the hash-chained index stays forever. Meets ISO 27001 A.8.15 and equivalent DORA record-keeping.
• Disaster recovery. RDS point-in-time recovery 7 days. Nightly age-encrypted database dumps to eu-nl. WORM audit anchor cross-region replica. Documented drill cadence.
• One-click tamper-evident export. Signed tarball: repos + LFS + Actions logs + audit slice + attestations + SHA-256 manifest. If you ever want to leave, standard formats throughout.

Public API for AI agents

Every admin action is a documented REST endpoint. AI coding agents operate against the same surface as the humans, on the same auth model.

frembench — your agent keeps working when your laptop closes

A Linux machine you rent by the minute and reach from the browser. Bring your own agent — Claude Code, Codex, opencode. €39/seat/month with 200 hours included.

€39 per seat per month includes 200 hours of bench runtime. Beyond that, €0,20/hour — and your bill can never exceed €99 per seat in a month however hard you run it. Hours pool across seats, like CI minutes. Full pricing · Documentation

Where it runs

Every automated processing surface is operated by entities with no US parent.

See trust for the full sub-processor list, inherited certifications (T Cloud Public: ISO 27001, 27017, 27018, BSI C5 Type 2, TISAX), the own-certification roadmap (ISO 27001 Stage 1 audit within 18 months, certification within 24 months, per DPA §12.1), and the DPA for the contractual posture.

What it isn't

Questions