Persistent Systems
  • Home
  • Platform
  • Proof
  • Research and Innovation
Issue 1

SASVA™ - INTELLIGENT ENTERPRISE AI

RESEARCH AND INNOVATION

From Model Capability to Enterprise System Science

Frontier models provide powerful intelligence, but they do not provide the complete system required to carry secure, long running, and accountable enterprise work. SASVA’s 140+ patents form an integrated intellectual property portfolio for secure, long running, and accountable enterprise execution focused on the capabilities required around the model to make enterprise execution possible.

SASVA combines secure and on-premises deployment, deterministic controls, governed execution, lower-cost model routing, persistent enterprise context, and the SASVA Brain. Its architect-first approach establishes intent, authority, dependencies, security conditions, cost limits, and evidence requirements before execution begins. Complex initiatives are then organized through controlled rounds and waves, with coordinated participation across agents, humans, tools, and enterprise systems.

The portfolio extends across executable architecture, governed knowledge, dynamic agent coordination, multi-dimensional memory, human intervention, self-correction, evidence-based assurance, sovereign execution, and cost control. These capabilities operate as one connected system: architecture shapes context, context guides agents, coordinated execution produces evidence, evidence strengthens memory, and memory improves subsequent execution. Together, they transform frontier-model capability into secure, deterministic, cost-efficient, and scalable enterprise outcomes.

Innovation Domain I: Executable Design and Governed Knowledge

SASVA makes human judgment computationally active throughout execution.

Executable architecture converts an architect’s storyboard into a machine-operable design, encoding phases, dependencies, decision rights, model and tool selection, security conditions, cost limits, evidence requirements, recovery paths, and human escalation. The SASVA Brain makes this design visible, shareable, replayable, and reusable.

Design-time intelligence enables domain experts to establish deterministic guardrails and validation logic before runtime. Proven migration methods, regulatory workflows, and quality strategies can therefore be captured as reusable Brains instead of being reconstructed through prompts.

SASVA also governs what the system knows and the authority assigned to that knowledge. Typed registries, provenance, lineage, staleness propagation, derivation chains, conflict priorities, consumption modes, and phase-specific budgets transform enterprise context into governed computation.

Primary context can be generated once and used to derive focused security, performance, testing, and onboarding artefacts. Only impacted content is refreshed, while complete lineage identifies the source behind every derivative. This foundation also supports context-impact analysis, quality scoring, cross-workspace federation, and adaptive consumption.

Innovation Domain II: Coordinated Agency and Evidence-Based Assurance

SASVA coordinates agents, people, tools, and environments as one governed execution system.

Dynamic orchestration constructs the execution topology from the actual structure of the work. Planning determines phases and dependencies, while the runtime creates specialist sessions for each task. Independent activities run in parallel, dependent activities remain sequenced, and budgets adapt to complexity. Continuation sessions, checkpoints, recovery mechanisms, and downstream propagation support reliable long-running execution.

Multi-dimensional memory preserves learning across sessions, corrections, teams, and projects. Human expertise is applied at the phases where it creates the greatest value, while live intervention allows agents to be paused, redirected, or constrained without losing execution continuity.

Assurance is built around evidence rather than task completion alone. Structural completeness, build integrity, functional equivalence, visual fidelity, testing, security, performance, scalability, accessibility, localization, and compliance can be assembled into the proof framework required for each system.

Self-correcting loops connect failures to the responsible agent. Circuit breakers stop unproductive machine iteration, while multi-environment execution validates outcomes across Windows, Linux, containers, remote systems, and cluster infrastructure within one governed workflow.

The result is autonomy that remains observable, interruptible, recoverable, and empirically validated.

Innovation Domain III: Cost, Sovereignty, and Future-Ready Security

SASVA makes advanced intelligence economically and operationally viable for enterprise use.

The platform combines frontier models, private models, purpose-built small models, deterministic engines, and design-time intelligence. Routing decisions account for complexity, sensitivity, latency, quality, and cost. Knowledge distillation and specialist models bring capability closer to the task, while persistent memory and derived context reduce repeated exploration. Deterministic transformation handles mechanical work where model reasoning adds limited value.

The storyboard brings cost architecture into pre-production. Context budgets, model selection, execution resources, and retry limits can be defined before consumption begins. Better decomposition also eliminates unnecessary inference cycles.

The same system supports enterprise sovereignty through private, on-premise, hybrid, and air-gapped execution; incognito operation; controlled synchronization; isolated runtimes; governed access; and complete auditability.

The roadmap extends this foundation into quantum-inspired optimization, post-quantum security, advanced reasoning, long-context execution, and scientific AI, with each area directed toward defined enterprise constraints and operating requirements.

An Integrated Intellectual Architecture

SASVA’s 140+ patents represent one connected system of invention.

The portfolio covers executable architectural intent, governed enterprise knowledge, dynamic agent coordination, persistent memory, failure recovery, functional equivalence, evidence-based assurance, private and purpose-built models, deterministic intelligence, advanced optimization, and future-ready security.

The value lies in how these inventions reinforce one another. Governed context improves agent decisions. Better orchestration reduces context pollution. Stronger verification produces better correction memory. Design-time intelligence and curated context improve the effectiveness of private and smaller models. Cost efficiency increases when the system distinguishes reasoning from deterministic transformation.

From Enterprise Constraint to Protected Innovation

SASVA’s innovation cycle begins in live enterprise environments, where technical constraints and their operational consequences can be directly observed.

Recurring constraints are abstracted into general engineering problems. Researchers and engineers develop mechanisms, algorithms, workflows, controls, and model techniques, then validate them as part of the wider SASVA system across context, orchestration, runtime, and assurance.

Patent protection captures the technical contribution. Productization integrates it into the platform. Enterprise execution validates it across large codebases, regulated data, multi-repository estates, sovereign environments, air-gapped infrastructure, and mission-critical systems.

Operational evidence then informs the next cycle of platform innovation through newly observed failure modes, correction patterns, performance characteristics, and cost profiles.

The cycle is continuous: enterprise constraint, technical abstraction, protected invention, platform capability, operational evidence, and continued innovation.

Model Cognition Lab

Model Cognition Lab, part of Persistent, is the research function behind SASVA's model layer. As open-weight models continue to close the benchmark gap with frontier systems, competitive advantage moves away from any single model toward the system built around it. The lab's mandate follows from that shift: make smaller, purpose-built models perform, on a defined enterprise task, at the level of the best available frontier model.

The lab rethinks distillation from first principles. Rather than simply shrinking a large model, it builds intelligence around the smaller one. Large models generate the scripts, tools, guardrails and workflows that become a deterministic layer, guiding small models step by step toward consistent, high-quality results. Depending on the use case, the lab draws on full fine-tuning, supervised fine-tuning, distillation, adapter tuning and reinforcement learning, and applies thinking distillation, transferring a larger model's reasoning path into a smaller model rather than its answers alone. A parallel technique, applied to mixture-of-experts models, removes the portions a given task does not need, reducing model footprint by 30 to 40 percent without a measurable loss of accuracy. That deterministic orchestration layer, the part that makes outputs auditable, reproducible and reversible, is what lets smaller, purpose-built models deliver enterprise-grade outcomes.

SASVA Core carries this process end to end. As the platform's automated model conditioning engine, the Core orchestrates data pipelines, generates training signals and delivers custom-conditioned models built for each enterprise. It evaluates each run against those that came before it, narrows its search using what earlier tasks have already established, and halts a run once further iteration is unlikely to improve the outcome, turning model conditioning into a repeatable engineering process rather than a manual research exercise. Those models run on private infrastructure, letting agentic systems operate on sensitive data with control and confidence, at a cost that scales with the business rather than against it.

Most of these techniques are patent protected. Combined with Mavericks, Persistent's AI-trained engineers, and the data and ML depth, Model Cognition Labs forms the core of what SASVA runs on beneath the platform layer described earlier.

Measuring this work required a different kind of benchmark. Public benchmarks are drawn from past pull requests in open repositories, and once published, the same tasks tend to surface in the next generation of pre-training data, a pattern researchers call task contamination. They also concentrate on single-repository, single-function problems, which understates enterprise work that spans hundreds of repositories, carries non-functional requirements, or requires a full technology migration rather than a local fix. SASVA's benchmark tasks are written by domain experts, not generated by other models, and its task set spans small, medium and complex problems, including changes that cross multiple repositories and full modernization efforts, scored on build correctness, security posture and code quality alongside test pass rate.

In a controlled benchmark spanning 189 runs, 32 migration tasks, and 12 production open-source repositories, SASVA achieved the highest full-suite score of 85.2 and was the only system to complete all 32 tasks. It slightly outperformed the leading frontier alternative while operating at approximately one-sixth of the cost, with code and data remaining within the enterprise environment.

The mission gives the portfolio its coherence: make advanced intelligence affordable enough to scale, reliable enough to carry consequential work and secure enough to operate inside the systems institutions protect most closely.

Every patent represents a sustained attempt to solve the foundational problems that appear when intelligence moves from assisting an individual to carrying an enterprise initiative. The portfolio establishes ways to make architectural intent computational, govern what agents know and use, coordinate many agents around one design, preserve memory across long-running work, make completion evidence-based, transform systems without losing business meaning, and operate advanced intelligence within private, secure and commercially disciplined boundaries.

This is the research foundation of SASVA: a connected science of design authority, governed knowledge, coordinated agency, verifiable execution and secure, affordable intelligence with dependable institutional capability.

Source: Persistent Systems
Return to Home
Engineered Inventions
140+
Patents. One connected system.

 

Benchmark Scorecard
85.2
Full-suite score
32/32
Tasks completed

189 runs · 12 production open-source repositories: highest score, only full completion.

 

Where SASVA Differentiates
Executable Design & Governed Knowledge Coordinated Agency & Assurance Cost, Sovereignty & Security

 

Definition

Thinking Distillation

Transferring a model's reasoning path, not just its final answer, into a smaller, purpose-built model.

 

Quotable
"Competitive advantage moves away from any single model toward the system built around it."
Gartner

SASVA™ - INTELLIGENT ENTERPRISE AI is published by Persistent Systems. Editorial content supplied by Persistent Systems is independent of Gartner analysis. All Gartner research is used with Gartner's permission, and was originally published as part of Gartner's syndicated research service available to all entitled Gartner clients. © 2026 Gartner, Inc. and/or its affiliates. All rights reserved. The use of Gartner research in this publication does not indicate Gartner's endorsement of Persistent Systems' products and/or strategies. Reproduction or distribution of this publication in any form without Gartner's prior written permission is forbidden. The information contained herein has been obtained from sources believed to be reliable. Gartner disclaims all warranties as to the accuracy, completeness or adequacy of such information. The opinions expressed herein are subject to change without notice. Although Gartner research may include a discussion of related legal issues, Gartner does not provide legal advice or services and its research should not be construed or used as such. Gartner is a public company, and its shareholders may include firms and funds that have financial interests in entities covered in Gartner research. Gartner's Board of Directors may include senior managers of these firms or funds. Gartner research is produced independently by its research organization without input or influence from these firms, funds or their managers. For further information on the independence and integrity of Gartner research, see "Guiding Principles on Independence and Objectivity", on its website.

About Gartner|Careers|Newsroom|Policies|Site Index|IT Glossary|Contact Gartner