Compact by design
A local AI reference capability intended for evaluation where footprint, power, and simplicity matter.
Expeditionary edge / compact deployment
Local AI support for forward units.
Mimir gives Army and Marine Corps teams local access to approved knowledge, cited troubleshooting, and guarded diagnostics—enhancing the people who maintain, plan, and sustain operations beyond reliable connectivity.
Topology
Compact 3-node
Connectivity
DDIL-oriented
AI posture
Local + source-grounded
Authority
Human
Designed for the forward edge
AI that travels with the mission—not the network.
Forward teams should not lose access to useful AI because a cloud path disappears. Mimir packages local models, governed knowledge, and an operator-ready interface into a system that can remain useful inside the unit’s own operational boundary.
Useful without reach-back
After services and content are staged locally, core retrieval and inference can operate without WAN reach-back.
Grounded in unit knowledge
Knowledge answers are designed to expose citations from unit-approved technical and operational material.
Built around the operator
Mimir reduces search burden while Marines, Soldiers, and leaders retain judgment and decision authority.
PacStar-class + compact OpenShift concept
A three-node compact cluster with local support context.
A compact OpenShift cluster places control-plane and workload roles across three nodes. Mimir can be evaluated alongside PacStar-class infrastructure as a local knowledge and troubleshooting node. This notional pattern has not yet been validated on PacStar hardware or through a live compact-cluster install.
Notional edge reference architecture
Compact OpenShift support pattern
Local boundary
Node 01
Control plane + compute
RHCOS managed
Cluster quorum
Workload capacity
Node 02
Control plane + compute
RHCOS managed
Cluster quorum
Workload capacity
Node 03
Control plane + compute
RHCOS managed
Cluster quorum
Workload capacity
Mimir appliance
RHEL 9 · Podman · local AI
Retrieval, reranking, citations, governed packs, and diagnostics sized to the available edge compute and power envelope.
Authorized forward team
Soldiers, Marines, and maintainers receive cited guidance and evidence while retaining approval and decision authority.
01
Define the disconnected chain
Document mirrored release content, registry trust, OpenShift tools, RHCOS media, checksums, DNS, NTP, and site-specific inputs for a future validated pattern.
02
Validate the compact pattern
Exercise topology, quorum-sensitive assumptions, agent hosts, rendezvous addressing, failure modes, and approval gates in a representative lab before field use.
03
Support day-two operations
Collect bounded, read-only cluster and node evidence, then pair it with approved troubleshooting guidance and citations—without turning AI into an unsupervised administrator.
Secure from the baseline up
Hardening planned and evidenced for the boundary.
We chose Red Hat Enterprise Linux because security controls belong in the platform—not in a slide deck. Mimir’s RHEL baseline, container runtime, application controls, and evidence plan are evaluated as parts of one customer-specific system boundary.
Post-quantum cryptography path
A scoped PQC enablement path.
RHEL 9.7 and later provide an optional PQ cryptographic subpolicy for supported components. The selected RHEL release, application path, peers, policy, compatibility, and evidence must be tested; Mimir is not currently certified as a PQC-compliant system.
Operating system
RHEL security foundation
Mimir is designed for a RHEL 9 baseline with SELinux enforcing, restricted service exposure, and a documented path for customer-specific scanning and STIG tailoring.
Runtime
Podman-contained services
Application, database, vector, and local-model services are packaged as a controlled container stack with explicit ports, local endpoints, and versioned deployment configuration.
Application
Roles, boundaries, and audit
Admin and operator roles, team-scoped connections, knowledge-pack isolation, local-only model routing, and recorded evidence keep access aligned to mission need.
Delivery
Evidence plan, not slogans
The deployment plan identifies configuration records, scan results, inventories, remediation notes, exceptions, and boundary-specific evidence to produce and review.
A configured RHEL host is a foundation—not an automatic ATO. Mimir is not FIPS 140 validated. Final STIG tailoring, application, database, container, network, and RMF evidence belong to the specific boundary.
Seedable starter-pack library
Useful before the first trouble ticket.
The repository includes 195 Mimir-authored troubleshooting cards that can be seeded into four starter-pack families. Each deployment must verify the cards actually loaded, indexed, reviewed, and approved. Customer manuals, SOPs, diagrams, and lessons learned can then be added through a governed release process.
195
Repository cards
4
Starter families
Local
Core answer path
101 cards
Red Hat platform
RHEL, OpenShift, Ansible Automation Platform, and Satellite operations.
- API and operator health
- DNS, SELinux, storage, and boot
- Disconnected install and lifecycle
26 cards
Cisco networking
Field-oriented switching, routing, firewall, and connectivity triage.
- Layer 1–3 isolation
- VLAN and routing checks
- Safe configuration triage
34 cards
Microsoft platform
Windows Server, identity, DNS, endpoint, and core service troubleshooting.
- Active Directory health
- DNS and authentication
- Service and endpoint triage
34 cards
VMware platform
vSphere, ESXi, vCenter, HA, networking, and storage first checks.
- Host and cluster health
- Datastore and path issues
- VM and service recovery
Mimir authored these cards. Vendor names identify technology coverage and do not imply vendor authorship, endorsement, certification, or replacement of current vendor documentation.
Review the pack model01
Situation framing
Classify the symptom before acting.
02
Safe first checks
Preserve evidence and narrow the fault.
03
Do-not-do-first
Warn against risky reflex actions.
04
Inspectable sources
Trace guidance back to approved material.
Forward mission applications
Put the reference library beside the team.
Mimir is most valuable where time, bandwidth, and access to specialized expertise are constrained.
Readiness
Maintenance at the point of work
Help maintainers move across technical manuals, inspection criteria, and troubleshooting procedures without waiting on a remote help desk.
- Fault isolation
- Inspection procedures
- Parts and maintenance context
Operations
Faster staff orientation
Give forward command elements a governed way to navigate doctrine, SOPs, orders, and reference libraries during planning and execution.
- Doctrine reference
- Planning support
- Watch-floor knowledge
Sustainment
Local logistics knowledge
Bring supply guidance, movement procedures, and sustainment references closer to the teams coordinating support in constrained environments.
- Supply workflows
- Movement references
- Policy and compliance
Notional deployment concept
A sovereign knowledge node at the tactical edge.
Approved content is prepared and governed by the mission owner, loaded onto the Mimir node, and made available to authorized users locally. The unit maintains control of its data and operating posture.
Final compute, storage, environmental qualification, power, and network interfaces must be selected and validated for the mission BOM and authorization boundary.
Forward knowledge flow
Customer-controlled from source to answer
Approved sources
TMs · SOPs · doctrine
Local Mimir node
Governed retrieval · inference
Authorized team
Cited answers · human action
Public model APINot required
Shared-model trainingNot required
Decision authorityUnit
The human advantage
More mental bandwidth for the work only people can do.
Mimir handles the search and source retrieval. The team applies experience, mission context, and judgment.