Parinita Instrument
The FIPS-Hardened Immutable OS Every Parinita Workload Boots On
12 factory images. Sub-12-minute bootstrap. 10-year CVE lifecycle. Immutable A/B partitions.
Instrument is a FIPS 140-3 hardened, DISA STIG-compliant Ubuntu 24.04 LTS image with kernel 6.18, augmented by Kairos for immutable A/B partition management. It ships in 12 factory images — one per silicon type — and is the only OS certified to run uniformly across this many silicon types simultaneously with zero-touch provisioning.
Heterogeneous AI Fleets Boot Inconsistent OS Images With No Measurement Chain
When different nodes in a multi-silicon AI fleet run different OS images with different security configurations, 'what was actually running on this node' cannot be answered with a measurement chain.
Generic enterprise Linux distributions are not designed for heterogeneous silicon, do not enforce immutability, and cannot prove what was running on a node at a given time with a tamper-proof record.
-
Multi-Silicon Fleets Have No Consistent OS Image
Different silicon types (MI350P, NVIDIA Blackwell, EPYC Turin, AmpereOne) typically run different OS configurations — inconsistency makes security controls non-uniform.
-
Generic Linux Has No Measurement Chain
Standard enterprise Linux distributions do not enforce immutability and cannot prove what was running on a node at a given time with a tamper-proof cryptographic record.
-
Ad-Hoc Package Installs Break Security Baselines
When operators install packages, modules, or patches outside the defined image, the security baseline breaks and the measurement chain is invalidated.
One immutable OS. Every silicon type. Cryptographic measurement chain.
Instrument's measurement chain runs from firmware through bootloader through kernel through initramfs into the TPM — every step recorded. What ran on which node at which time is cryptographically verifiable, not screenshot-based.
Arbitrary kernel modules, custom firmware, and ad-hoc package installs are policy-blocked by design — that is what makes the measurement chain meaningful.
-
12 Factory Images
One image per silicon type — AMD Instinct MI350P, NVIDIA Blackwell, AMD EPYC Turin, Sierra Forest, Qualcomm IQ-9075M, AmpereOne ARM64, AMD Alveo FPGA, NVMe storage, Qualcomm Cloud AI 100 edge, video, orchestration — all Cosign-signed.
-
Sub-12-Minute Bootstrap
PXE boot, FIPS-mode activation, measured-boot via TPM, Maestro registration, fabric join — power-on to workload-ready in under 12 minutes across all 12 silicon types.
-
Immutable A/B Partitions
New images write to the inactive partition, validate, flip the boot target, and auto-rollback on any boot or health failure. Every push anchors on Chrysalis.
-
TPM Measurement Chain
Measured-boot chains firmware, bootloader, kernel, and initramfs into the TPM — every step recorded. What ran on which node at which time is cryptographically verifiable.
-
FIPS 140-3 and DISA STIG Compliance
Ubuntu 24.04 LTS with kernel 6.18, FIPS mode activated at first boot, pre-applied DISA STIG hardening — compliance is factory-baked, not post-deployment configured.
-
10-Year CVE Lifecycle
Long-term security maintenance for kernel, userspace, and silicon drivers on a Parinita-owned cadence — independent of upstream distribution lifecycle.
-
Pre-Hardened at Boot
Encrypted root and data partitions, secure-boot certificates, Wazuh agent factory-integrated, ConnectX-7 drivers with eBPF/XDP, hardware IPsec/AES-GCM crypto — all present at first boot.
-
Chrysalis-Anchored Image Pushes
Every image push records on Chrysalis as a signed artifact lineage record — cryptographic proof of what image was pushed to which node, when, by whom.
-
Zero-Touch Provisioning
Factory image at the manufacturing line or PXE push at the POP. No manual OS configuration step. The runtime agent registers automatically with Maestro on first boot.
The Plane Model
Orchestra introduces "planes" — logical groupings of hardware optimized for a specific workload class. Unlike Kubernetes node pools, planes represent fundamentally different hardware architectures with different drivers, network requirements, and scheduling semantics.
The plane model is what makes Orchestra different from every other orchestration tool. Kubernetes sees nodes. Orchestra sees purpose-built hardware tiers and routes workloads accordingly.
Cosign-signed image. TPM measurement chain. Under 12 minutes.
PXE boot or factory image, FIPS mode activation, measured-boot chain from firmware to initramfs via TPM, registration with Maestro, fabric join — power-on to workload-ready in under 12 minutes.
- 01Silicon type12 factory images, one per silicon — MI350P, Blackwell, EPYC Turin, Sierra Forest, IQ-9075M, AmpereOne, Alveo FPGA, NVMe, Cloud AI 100, video, orchestration.
- 02Boot pathPXE boot or factory pre-flash at manufacturing line. FIPS mode activates at first boot. TPM measurement chain records every boot step.
- 03Partition modelImmutable A/B partitions via Kairos — new image writes to inactive partition, validates, flips boot target. Auto-rollback on any boot or health failure.
- 04Measurement chainFirmware through bootloader through kernel through initramfs — each step extends into the TPM. The full chain is cryptographically verifiable and anchors on Chrysalis.
- 05Policy enforcementArbitrary kernel modules, custom firmware, and ad-hoc package installs are policy-blocked. The measurement chain is only meaningful if the baseline is enforced.
- 06RegistrationRuntime agent registers with Maestro on first boot — silicon ID, plane type, POP location, FIPS status reported automatically. Bootstrap completes in under 12 minutes.
- 07CVE maintenance10-year security maintenance cadence for kernel, userspace, and silicon drivers — independent of upstream Ubuntu or silicon vendor lifecycle.
- 08Image signingAll 12 factory images Cosign-signed. Every push to a node records on Chrysalis as a signed artifact lineage record — what ran where and when is independently verifiable.
import instrument
client = instrument.Client(seat_token='your_seat_token')
# List all nodes with their image version and FIPS status
for node in client.nodes.list(pop_id='ord-t1-01'):
print(f"Node: {node.id}, Silicon: {node.silicon_type}, Image: {node.image_version}, FIPS: {node.fips_active}")
# Query TPM measurement chain for a specific node
chain = client.nodes.measurement_chain(node_id='node-mi350p-042')
for step in chain.steps:
print(f"Step: {step.name}, Hash: {step.hash}, Chrysalis: {step.chrysalis_tx}")Proven at scale. Not in a lab.
Parinita AI Edge is the production deployment of the Parinita platform and the largest heterogeneous AI infrastructure deployment in the United States.
Parinita AI Edge
The most complex heterogeneous AI infrastructure in the United States. 101 sites, 9 planes, 12,000+ nodes, 4 accelerator vendors, dual network fabrics, four-layer tenant isolation — all through a single sovereign control plane.
Network & Security Infrastructure
- Multi-vendor acceleratorsFour accelerator vendors — Intel Habana, NVIDIA, AMD, Qualcomm — orchestrated through one control plane with unified scheduling, monitoring, and lifecycle management.
- Dual-fabric networkingCisco production fabric and Arista GPU backend fabric operating as a coordinated system, bridged by identity-aware routing.
- Nationwide scale101 sites across 42 U.S. states, each operating autonomously with a local control agent and a sovereign cross-site routing plane.
- Multi-tenant isolationFour-layer defense-in-depth: VXLAN VNIs, identity-routing, Palo Alto firewalls, and Cilium eBPF — validated across every plane and site.
- Compliance readinessFIPS 140-2 at launch, with FedRAMP Moderate, CJIS, and IL4/IL5 certification paths active through Parinita compliance profiles.
- Sub-millisecond routingEvery request classified and dispatched in under 1ms, enabling real-time SLA enforcement without perceptible overhead.
For deployments where 'what was running' must be answered with a measurement, not a screenshot.
Instrument is most valuable when multi-silicon consistency is a hard requirement and the measurement chain has regulatory or legal significance.
Not a general-purpose distribution — arbitrary kernel modules and ad-hoc package installs are policy-blocked. That constraint is what makes the measurement chain trustworthy.
-
Multi-Silicon AI Infrastructure
Instrument is the only OS certified to run uniformly across MI350P, Blackwell, EPYC Turin, AmpereOne, and Qualcomm silicon simultaneously — one consistent security baseline.
-
Regulated and Audited Environments
Deployments where 'what was actually running on this node' needs a measurement chain answer — not a screenshot, not a log, but a TPM-recorded cryptographic proof.
-
Large-Scale Edge Fleets
Zero-touch provisioning via PXE or factory pre-flash scales across 12,000+ nodes without manual OS configuration. Sub-12-minute bootstrap from power-on to workload-ready.
-
Long-Lived Infrastructure
10-year CVE lifecycle on kernel, userspace, and silicon drivers — infrastructure deployed today stays maintained for the decade without upstream distribution lifecycle risk.
-
FIPS and STIG Compliance
FIPS 140-3 mode and DISA STIG hardening factory-baked at first boot — compliance is architectural, not a post-deployment configuration project.
Deployment Models
Factory images provision at the Supermicro, Dell, or Qualcomm Cloud AI 100 manufacturing line or pushed over PXE at the POP.
Factory images apply at the Supermicro, Dell, or Qualcomm Cloud AI 100 manufacturing line. Nodes arrive at the POP pre-configured — PXE not required.
PXE boot pushes the correct silicon-specific factory image at the POP. FIPS activation, measured boot, and Maestro registration complete automatically. Under 12 minutes total.
New Instrument image versions roll out via Opera's coordinated canary, staged, and fleet deployment process — with automatic rollback on attestation failure or SLO breach.
Talk to Us
Our infrastructure team can walk through Instrument's factory image selection and bootstrap process for your specific silicon and POP configuration.