Citadel
VERSION 1.0.0 · GENERALLY AVAILABLELive host-state evidence for regulated fleets.
Citadel is a host-state evidence platform for regulated environments. It deploys and configures agents across a host fleet and turns their results into a signed, queryable evidence stream, so "what is actually running on every host" is a SQL question with a timestamped answer.
Citadel —What it is built to do
Fleet-scale collection over mutually authenticated TLS
Enrolment is an installer and a config file; packs are distributed per platform from the control plane; every result is signed at the agent into an append-only store you own. Verified 20 August 2026: a control plane behind a real certificate chain with a macOS host enrolled, online, and returning its scheduled results over TLS.
Closed beta · running in production at Novaprospect
Request beta access →The problem
Compliance evidence is a snapshot of a thing that moves.
Traditional endpoint evidence is batch: quarterly scans and CMDB reconciliation. Between collections the boundary drifts, and the drift surfaces at the next scan or the next audit.
An on-host agent solves the telemetry side: every host becomes a SQL-queryable surface exposing hundreds of tables of live state. An agent alone does not handle fleet-scale operation — deployment, configuration management, query distribution, result aggregation, evidence retention, and the compliance mapping an assessor needs.
Citadel is that layer.
How it works
Four pillars of fleet-scale host evidence.
Each pillar produces a concrete artifact the platform generates and retains.
Fleet-scale agent orchestration
Designed to enrol, version, and configure the Citadel agent across thousands of hosts — Linux, macOS, Windows, and containers. Configuration is policy-driven and served from the control plane.
Compliance-aligned query packs
Curated query packs map to NIST 800-53 control families — CM-6 (config settings), CM-8 (component inventory), SI-4 (system monitoring), SI-7 (software integrity), AU-2 (auditing). Evidence is produced as a byproduct of operation.
Signed, queryable evidence
Every result is timestamped and signed at the agent. Historical state is retained in an append-only store — so assessors can answer "what did host X look like on date Y" with a SQL query.
Air-gap and FIPS-capable
Deploys in air-gapped environments with no outbound dependencies. FIPS 140-3 agent builds are available for regulated workloads.
Capability
What it is built to do, and what we have verified.
Each claim below is labelled with its basis. Measured means we ran it and checked it. Designed means it is what the architecture supports.
A control plane behind a real certificate chain, with a host enrolled and reporting
Verified 20 August 2026: a Citadel control plane running behind a genuine certificate chain, with a macOS host enrolled and ONLINE, running four scheduled query packs — 38 scheduled queries — and returning results over TLS. One host.
Built for fleet-scale enrolment
Agent transport is osquery-compatible TLS: enrolment is an installer and a config file, and configuration is served from the control plane. Adding hosts is intended to be a deployment problem.
Packs distributed per platform
A pack declares per-platform overrides for Linux, macOS and Windows, so one reviewed pack covers a mixed estate rather than forking into three. Packs are YAML, held in Git, and released signed.
Append-only evidence, queryable after the fact
"What did host X look like on the day of the finding" is a SQL question against a retained, signed store — Postgres or ClickHouse, customer-owned. Retention is sized to the authorization period.
Air-gap and FIPS 140-3 capable
Control plane, agents and evidence store run with no outbound dependency, and FIPS 140-3 agent builds are available for regulated workloads.
Ingests an osquery fleet you already run
Raw osquery results can report into the same evidence store, so an existing fleet becomes evidence without a re-enrolment programme first.
Integrations
Feeds your SIEM, your SSP, and your ticket queue.
Citadel is built around the assumption that the data it produces is only useful if it lands in the systems your team already uses for evidence, detection, and remediation.
Query packs & scheduling
Query packs are authored as YAML, reviewed in a pull request like any other change, and promoted through environments. Each pack is mapped to the 800-53 controls it produces evidence for — so nothing is orphaned.
- —Git-backed pack repository with signed releases.
- —Ad-hoc queries scoped by host label or policy group.
- —Scheduled packs with per-platform overrides (Linux, macOS, Windows).
- —Curated baselines aligned to CIS, NIST 800-53, and FedRAMP ConMon.
- —Ingest of raw osquery results, so a fleet you already run can report into the same evidence store.
Result stream
Results land in an append-only store and forward to the customer SIEM in real time. Every row is signed by the agent that produced it and linked to the pack, query, and host that produced it.
- —SIEM targets — Splunk, Elastic, Chronicle, Sentinel, QRadar.
- —Historical SQL interface over the evidence store.
- —Anomaly detection on result deltas with optional ticketing hooks.
- —Native export to the FedRAMP Management Engine as OSCAL origin references.
Paired with the platform
Host truth feeds the SSP — and verifies AI work.
Citadel gives the FedRAMP Management Engine a live, signed answer to the question every CM-family control asks: "what is actually on these hosts right now?" And it closes the loop on NAICOM-generated changes by verifying that a commit's intended effect actually landed on every target host.
Inventory from ground truth
CM-8 evidence comes from live agent results. The SSP cites the query, the host set, and the timestamp.
Change verification
After a NAICOM-tracked change lands, a deploy-verification pack re-runs against the target hosts to confirm the expected state.
Drift becomes POA&M
Unexpected deltas in a scheduled pack auto-open POA&M candidates in the Engine — with the host, query, and delta pre-filled.
Continuous ConMon
Replaces quarterly screenshots with a signed event stream that satisfies FedRAMP continuous-monitoring expectations.
Compliance alignment
Control-family evidence produced as a byproduct of operation.
Citadel query packs are authored to produce evidence mapped to specific 800-53 controls. No separate evidence-gathering pass at assessment time.
NIST 800-53 — CM family
Produces CM-2 (baseline configuration), CM-6 (configuration settings), CM-7 (least functionality), and CM-8 (component inventory) evidence on a schedule, live from every enrolled host.
NIST 800-53 — SI family
SI-4 (system monitoring) and SI-7 (software integrity) evidence produced continuously. File-integrity queries, process monitoring, and package-hash verification run as scheduled query packs.
FedRAMP ConMon
Replaces quarterly scan screenshots with a live, queryable evidence stream. Satisfies the continuous-monitoring requirements FedRAMP 20x emphasizes.
Architecture
Control plane inside your boundary. Agents everywhere.
Citadel runs as a self-hosted control plane — agent registry, query distributor, result collector, and evidence store — inside the customer's authorization boundary. On each host the Citadel agent handles policy-controlled scheduling and signed result emission, and is versioned and configured from the control plane.
Results land in a customer-owned append-only store (Postgres or ClickHouse) and forward to the customer's SIEM over mutually authenticated TLS. The Engine reads from the same store for OSCAL origin references.
Air-gapped deployments are supported natively — the control plane, agents, and evidence store run without any outbound dependency.
Who this is for
Four things that make Citadel worth your time.
These are not gates — if your situation is close but not exactly this, the conversation is still worth having.
Hosts whose live state you cannot answer for
Today "what is running right now" comes from a quarterly scan or a CMDB that drifts. Starting from nothing is fine — enrolment is an installer and a config file.
Endpoints to enrol
A real fleet of Linux, macOS, or Windows hosts inside a boundary you control. Tens is enough to be useful; the interesting questions start in the hundreds.
A compliance or asset-inventory driver
CM-2, CM-6, CM-8, SI-4 or SI-7 evidence you currently assemble by hand, a CMDB that drifts from reality, or FedRAMP continuous monitoring you are trying to move off a screenshot cadence.
Somewhere for the results to land
A SIEM, a data warehouse, or a willingness to run the append-only evidence store we ship. Air-gapped deployments are supported natively.
What the conversation covers
A live control plane.
- A control plane with hosts enrolled, live — not slides, and not a recorded video.
- Enrolment end to end: an installer and a config file, reporting over TLS.
- The query packs and the 800-53 controls each one evidences, and how a pack is reviewed and released.
- A historical query — what a host looked like on a date that has already passed.
- A deployment plan for your own boundary, including the air-gapped path if that is yours.
What we need from you
Thirty minutes and a named contact.
- A named technical contact who can speak for the fleet Citadel would run against.
- A rough shape of that fleet — host count and operating systems, and whether any of it is air-gapped.
- Where you want results to land: a SIEM, a warehouse, or the evidence store we ship.
- The deadline actually driving this: an authorization date, a ConMon obligation, an audit, or a customer security review.
FAQ
What people usually ask first.
What does the first conversation involve?
A control plane with hosts enrolled, live. We show enrolment (an installer and a config file, reporting over TLS), the query packs and the 800-53 controls each one evidences, and a historical query — what a host looked like on a date that has already passed. Air-gapped deployments are supported natively and we will walk that path if it is yours.
Is Citadel released?
Yes. Citadel is at version 1.0.0 and generally available — control plane, agent, and compliance query packs. It is deployed inside your own boundary rather than accessed as a hosted service.
How large a fleet can it handle?
Host count is a control-plane sizing question. Agent transport is the osquery TLS remote API — the same transport osquery fleets run at scale today — and configuration is served centrally rather than per host, so hosts add load to the control plane and nothing else. Collection interval and pack volume drive that load more than raw host count does, and we size the control plane against your schedule and your fleet when we talk.
Can we use the osquery agents we already run?
Yes. Raw osquery results can be ingested into the same evidence store, so an existing fleet can start producing retained, queryable evidence before any re-enrolment programme.
Does it work air-gapped?
Natively. The control plane, the agents and the evidence store run with no outbound dependency, and FIPS 140-3 agent builds are available for regulated workloads.
How is it available?
As a closed beta. Citadel runs in production at Novaprospect against its own host fleet, and access goes to organizations that ask for it and say what they would run it against. Request beta access from this page and you get an answer. No rate is published for it. Acquisition conversations are open separately; the acquisition page sets out what a buyer inherits.
Know what's on every host.
Citadel 1.0.0 is released and deployed inside your own boundary — for regulated fleets under FedRAMP, DoD IL, or equivalent continuous-monitoring regimes.
Closed beta · running in production at Novaprospect
Request beta access →