The Sovereign AI Operating Layer for Defence, National Security and Critical Systems
Tosh Defence connects data, models, agents, communications and mission applications inside infrastructure controlled entirely by the sovereign operator. Designed for on-premises, air-gapped and disconnected environments, with human authority, data control and auditability kept with you.
Customer control is the product.
The Operating Layer
One Sovereign Foundation. Many Mission Applications.
Tosh Defence is a sovereign defence AI and mission software company. The operating layer is the shared foundation that governs how approved data, models, agents, tools, identities and policies work together inside your security boundary. Our mission applications run on it, so the first accredited footprint becomes the base for every capability that follows.
Sovereign Deployment
Customer-operated on-premises, air-gap capable and sovereign-cloud deployment, with your identity, keys and update approvals.
Data and Integration Fabric
Canonical data models and adapters over heterogeneous systems, with lineage, provenance and access policy.
PRAGYA Intelligence
Model-flexible AI on replaceable foundations, selected per task, language, hardware and security boundary.
YANTRA Orchestration
Agentic workflows with allow-listed tools, visible steps, human approval for consequential actions and attributable audit.
Products
Six Mission Applications
Each application runs on, contributes to or extends the same sovereign operating layer: one secure footprint, with every additional capability a controlled extension of it.
SANKET
Sovereign secure information exchange platform for military units and high-trust organisations, deployed and governed inside the customer's own boundary.
- Signal Protocol end-to-end encryption through official libsignal bindings
- Organisational hierarchy and administrative governance
- Customer-operated, on-premises and air-gap capable deployment
RecordsKeeper.AI
AI-first records, knowledge and compliance intelligence that turns organisational records into a searchable, governed and auditable knowledge system.
- Ingestion, classification and natural-language retrieval
- Tamper-evident audit trails and chain of custody
- Continuous compliance evidence inside your boundary
Tosh AI
Sovereign AI, retrieval and agentic orchestration: the controlled reasoning and action system around replaceable models, running inside your own walls.
- PRAGYA model family on replaceable foundations
- YANTRA orchestration with allow-listed tools and human approval
- Retrieval with citations and an attributable audit trail
SAJAG
Predictive maintenance and spares-demand intelligence that estimates failure risk from historical maintenance, usage, terrain and weather data.
- Data-first, sensor-optional failure risk estimation
- Spares-demand forecasting from maintenance history
- Interpretable models that run on CPU-capable hardware
DRISHYA
Multi-OEM infrastructure monitoring that brings telemetry and alerts from heterogeneous network equipment into one operational picture.
- Unified Device Data Model with a vendor adapter framework
- Correlation of related alerts across vendors
- Device and protocol coverage confirmed per deployment
MAYA
Defensive cyber deception and incident intelligence: approved honeypots, honeytokens and canary artefacts that expose intrusions and give operators context.
- Approved honeypots, honeytokens and canary artefacts
- Observed behaviour mapped to MITRE ATT&CK
- Policy-bounded response under human authority
CAPABILITY PILLARS
Trusted communication - sensitive information is exchanged under customer control
Knowledge and compliance - records become searchable, attributable and useful
Mission reasoning - operators reason across approved data and tools inside the enclave
Asset readiness - equipment failures and spares demand are anticipated
Infrastructure awareness - heterogeneous devices are understood in one picture
Cyber resilience - intrusions are detected through deception and understood quickly
SAKSHAM Sensor-agnostic physiological readiness decision support. A pipeline specification, not a shipped product: no diagnosis, treatment or clinical claim.
Read the specification
Our Doctrine
How We Build
Customer Control Is the Product
Data residency is necessary but not enough. You control infrastructure, identity, keys, policies, updates and continuity.
Own the Intelligence, Integrate the Hardware
We own the data model, adapters, intelligence and operator workflow. Approved hardware and models stay replaceable.
Human Authority Stays Explicit
AI detects, retrieves, correlates, drafts and recommends. Consequential actions follow a defined authority model and human approval.
Disconnection Is Normal
Every application defines how it behaves when degraded, denied, intermittent or disconnected. Offline behaviour is designed and tested.
Evidence Outranks Adjectives
Architecture, test results, limitations and demonstrated behaviour, not superlatives. We would rather state a limit than inflate a promise.
Defensive Capability Only
Defensive, dual-use and decision-support software. No weapons, munitions, target selection, fire control or autonomous lethal decisions.
Open Interfaces Prevent Lock-In
Documented APIs, adapters, export paths and standard protocols, so you can inspect, migrate and keep operating.
One Footprint, Many Applications
A common deployment, identity, audit, policy and orchestration foundation. Each new application extends a footprint you already trust.
CONTROLLED DEMONSTRATION
See the Operating Layer in Action
A controlled, offline-capable demonstration scoped to your operational problem, security boundary and existing systems, with the evidence and limitations alongside every claim.
WHAT SOVEREIGNTY MEANS
Sovereignty Is an Operational Property
Sovereignty is not a flag or a hosting-region label. A system is sovereign only when the authorised operator controls every link in this chain, and we design each mission application to demonstrate it.
Where It Runs
Your premises, your air-gapped enclave or an approved sovereign cloud, chosen by you.
Where Data Lives
Data, keys, logs and model artefacts stay inside the boundary you define.
Who Administers It
You decide which administrators can reach each layer, through your own identity systems.
Which Models and Tools
Only the models and tools you approve are available to people and agents.
What Leaves the Boundary
No mandatory telemetry or call-home. Egress is your decision, not ours.
How Updates Arrive
Updates are inspected, approved and installed on your schedule, including offline.
How It Behaves Disconnected
Core workflows are designed to continue under limited or absent connectivity.
Who Did What
Every consequential action by a person or an agent is attributed and reviewable.
Continuity Without Us
The system is designed to keep operating without dependence on our live infrastructure.
Start with a Bounded Evaluation
Tell us the operational problem, the users, the security boundary and the systems involved. We will scope a controlled demonstration with clear acceptance criteria, for sovereign and approved organisations worldwide.