qf eBPF host firewall
security by design

Security by design: PKI, signed policies, mTLS

A PKI under the control plane, Ed25519-signed policy bundles, mTLS between agents and CP, and no phone-home. Self-hosted, air-gap capable — keys and data stay inside your perimeter.

what "secure by design" means

Trust is built into the chain, not bolted on

qf security rests on cryptography, not on trusting the network. The control plane runs its own certificate authority, signs every policy bundle with an Ed25519 key, and agents talk to the CP only over mTLS. Neither the agent nor the control plane reaches outward — all exchange stays inside your perimeter.

An agent applies a bundle only after verifying its signature: a policy that did not come from your CP, or was altered in transit, will not run. The Ed25519 signature attests to policy authenticity and integrity — it is not a licensing mechanism or a lock-in.

chain of trust

Four pillars of qf security

Identity, signature, channel, and no external dependencies — each verified cryptographically.

01 · PKI

A PKI under the control plane

The CP runs its own certificate authority and issues agent certificates at enrollment. Certificate revocation and enrollment tokens stay under your control. No external CAs, no cloud trust services.

02 · Ed25519

Signed policy bundles

Every bundle is signed with an Ed25519 key. An agent enforces a policy only after verifying the signature, guaranteeing authenticity and integrity. Ed25519 attests to the policy source — it is not a license.

03 · mTLS

Mutually authenticated channel

Agents and the control plane connect over mTLS gRPC: both sides present certificates. Bundles, counters, and events travel one authenticated channel — no inbound management port exposed on the agent.

04 · no phone-home

No outbound calls

Neither the agent nor the CP calls external servers — not for activation, not for telemetry. Enrollment via an out-of-band CA-fingerprint pin, an offline cache, and full air-gap operation. Keys and data stay in your perimeter.

no big-bang cutover

Microsegmentation rolls out in observable stages

Zero-trust segmentation fails on the risk of a single switch, not on the technique. qf separates observing from denying in time.

  1. 01

    Log — see the real picture

    Policies run in observe mode: qf logs what traffic passes between hosts without dropping anything. You see actual flows instead of guessing at topology.

  2. 02

    Observe — confirm intent

    Dry-run and preview show which hosts and rules a new policy will change before it applies. Observed traffic on the host page confirms nothing needed will be caught by a deny.

  3. 03

    Deny — enable segment by segment

    Enforcement is turned on one segment at a time, not fleet-wide at once. The cascade carries each change to all affected hosts, and audit records before/after at every step.

Fail-open by design — a rollout will not cause an outage on its own

By default qf runs fail-open (ALLOW): a forgotten or not-yet-rolled-out rule will not cut a host off. This is a managed, staged path toward default-deny, not "default-deny out of the box" — segmentation is enabled deliberately, not with one switch.

market context

Why microsegmentation is rated essential yet rarely deployed

Industry estimates put roughly 70% of organizations treating microsegmentation as important for zero trust, while only around 5% actually segment their network, with implementation complexity cited as the top barrier by up to 40% [source: zeronetworks.com, scworld.com]. The gap is not in the idea — it is in rollout risk and lack of visibility.

qf closes exactly that gap: host segmentation with no new hardware, a staged log → observe → deny path instead of a big-bang default-deny, and visibility into real traffic before you apply. These industry figures are external market context, not qf metrics.

self-hosted

Keys and data stay with you

The control plane is autonomous: state lives in your PostgreSQL and a PKI directory, with no Kubernetes, no etcd, and no external services. Private CA keys and policy-signing keys are generated and kept on your side. qf deploys on-prem, up to a fully air-gapped environment with no internet access.

In air-gap mode agents enroll via an out-of-band CA-fingerprint pin and update from a registry mirror, while an offline cache keeps enforcement running when the CP is unreachable. Nothing leaves the perimeter — not telemetry, not keys, not policies.

questions

Frequently asked

What exactly is signed, and with what?

Every policy bundle is signed with an Ed25519 key on the control-plane side. The agent verifies the signature before applying and will not enforce a bundle with a missing or invalid signature. The signature attests to policy authenticity and integrity — it is not a license key.

Does qf reach out to external servers?

No. Neither the agent nor the control plane phones home — not for activation, not for telemetry. Agents can enroll via an out-of-band CA-fingerprint pin and update from a local registry mirror. qf runs fully air-gapped with no internet access.

Is qf "default-deny out of the box"?

No. By default qf runs fail-open (ALLOW): a forgotten rule will not cut a host off. The move to default-deny is managed and staged — log, observe, deny, segment by segment, with before/after audit and a preview of affected hosts.

Where are keys and data stored?

On your side. Private CA keys and policy signatures are generated and kept in the control plane PKI directory, and state lives in your PostgreSQL. There are no external CAs or cloud trust services; qf deploys on-prem, including air-gapped.

Validate the security model on your own fleet