EvoMind
Flagship cognitive architecture for reasoning, memory, planning, governance, desktop operation, validation, and controlled improvement.
Independent applied AI systems lab
Building the operating layer for governed AI work.
SALT19 is the home of EvoMind — a flagship cognitive runtime for reasoning, memory, planning, validation, and local desktop execution — plus focused products that prove the architecture in real workflows.
PerceiveReasonPlanAct + Learn
Flagship system
EvoMind is the cognitive runtime at the center of SALT19.
EvoMind is SALT19’s core research and engineering platform: a governed runtime that organizes perception, memory, world modeling, reasoning, planning, validation, execution, observation, and controlled improvement into one operating loop.
Reasoning before actionTasks are decomposed, checked, and routed through policy-aware execution boundaries.
Memory-centered operationUseful task context, incidents, repairs, and strategies can persist beyond a single prompt.
Desktop-native directionThe architecture is aimed at real browser, document, file, and local app workflows.
Human-gated autonomyRisky handoffs, approvals, external routing, and source changes remain reviewable.
Evidence-first engineeringResearch, validation, traces, and benchmark work are separated from homepage marketing.
Product-backed lab workSignalLoom, Patriot Claim Tracker, and Recruiter.ai turn the doctrine into usable tools.
Cognitive architecture
Capabilities that compound into agentic work.
SALT19 presents the architecture primitives that matter for local-first autonomy: perceive, remember, model, reason, plan, govern, execute, observe, and improve.
Perception
Read desktop, document, browser, and application state.
Memory
Retain useful task context instead of starting cold every time.
World model
Track goals, constraints, entities, files, and workflow state.
Reasoning
Evaluate options before generating output or taking action.
Planning
Decompose work into steps with validation points.
Governance
Apply safety, scope, audit, and human-gate boundaries.
Execution
Operate local tools through controlled action layers.
Learning
Reuse incidents, repairs, traces, and strategies under control.
Product portfolio
Focused products built from the same SALT19 doctrine.
Each product applies the same governed runtime to a real workflow—turning the core architecture into usable systems for claims, invoices, job search, and operational automation.
SignalLoom
Local-first, human-gated invoice workflow agent that turns invoices and workbooks into extraction records, AP packets, approval text, exception checklists, and routing payloads.
Patriot Claim Tracker
Chrome tool for organizing VA.gov claims, messages, documents, benefits, payments, health, and forms into a cleaner local-first dashboard.
Recruiter.ai
Resume-first job matching assistant for exact listings, freshness signals, ghost-job risk, match scoring, and direct apply links.
Why local-first matters
AI work should not require surrendering the workspace.
SALT19’s direction is device-native and privacy-aware: keep files, workflow context, review packets, and sensitive decisions close to the user where practical. Integrations are useful, but autonomy should remain bounded, inspectable, and human-gated.
Try SignalLoomGovernance and safety
Control is part of the architecture, not a footer disclaimer.
EvoMind and SALT19 products are designed around explicit review, audit context, source preservation, and clear boundaries for risky actions such as payments, source mutation, external routing, and irreversible changes.
Open bench / validationResearch and proof
Published research and shipped systems.
SALT19 focus is on mission, capabilities, products, and research direction, with deeper engineering evidence available through the dedicated validation record.
EvoMind white paper
EvoMind is documented as a governed cognitive architecture for autonomous reasoning, planning, memory, desktop agents, validation, and controlled improvement.
Zenodo recordEngineering validation record
Desktop proofs, memory tests, governance checks, and benchmark work remain part of the engineering record. The validation page carries the deeper technical detail for diligence review.
Bench / ValidationLatest updates
What SALT19 is building now.
Built for real use, not concept demos. SALT19 is focused on deployable cognitive runtime systems, applied workflow tools, and evidence-backed validation.
EvoMind runtime
Advancing governed cognition, desktop execution, memory, planning, and validated improvement.
SignalLoom InvoiceOps
Local-first invoice review agent with Qwen/Ollama support, AP packets, approval text, and human-gated routing.
Applied workflow tools
Patriot Claim Tracker and Recruiter.ai extend the SALT19 doctrine into veteran workflows and job search automation.
FAQ
Clear answers for people and AI search.
SALT19’s focus: local AI agents, governed desktop runtime, and human-gated workflow automation.
What is SALT19?
SALT19 is an independent applied AI systems lab building EvoMind and local-first tools for governed cognition, real-world digital workflows, and accountable automation.
What is EvoMind?
EvoMind is SALT19’s flagship governed cognitive runtime for reasoning, memory, planning, validation, local desktop execution, observation, and controlled improvement.
What makes SALT19 different?
SALT19 focuses on local-first, device-native AI work, human gates, auditability, and practical products instead of cloud-only chatbot experiences.
Where is the technical validation?
Technical validation details are kept on the dedicated validation page so the homepage can stay focused on mission, capabilities, products, and research direction.
Frontier direction. Local execution. Human control.
SALT19 is building toward governed generality through real desktop workflows, persistent memory, planning, validation, and practical products that users can inspect.