Governed cognitive systems
for real-world work.
SALT19 researches and builds systems that reason over persistent state, compose capabilities, and verify outcomes. EvoMind is our flagship cognitive architecture.
EvoMind is SALT19’s governed cognitive architecture.
EvoMind is engineered as a persistent cognitive system rather than a prompt wrapper: reasoning and planning operate over durable state; semantic work requirements resolve into existing capabilities; execution passes through governed authority; and successful experience is reused only through controlled, reversible mechanisms.
Decomposition
Bounded validation exercises objective decomposition into distinct work-product requirements.
Verification
Distinguishes attempted operations from identifiable resulting artifacts with provenance.
Governed Execution
Planning does not automatically grant permission to act. Consequential execution passes through explicit admission and control boundaries.
A control architecture for cognition & evidence.
Organized around persistent state, reasoning and planning, semantic capability resolution, governed admission, real execution, and outcome verification.
Persistent State
Carry session, memory, project, and workflow context across more than one turn or process boundary.
Reasoning + Planning
Interpret desired outcomes, evaluate constraints, and construct multi-step capability plans before acting.
Capability Resolution
Express work semantically and bind requirements to available implementations rather than hard-coding paths.
Governed Admission
Separate cognitive intent from execution authority with explicit policy, scope, approval, and revocation boundaries.
Desktop Execution
Operate browsers, documents, files, and local applications through controlled execution layers.
Outcome Verification
Check resulting state rather than treating a returned tool call or attempted action as proof of success.
Artifact Lineage
Preserve identity, versions, source references, provenance, and relationships among resulting work products.
Experience Reuse
Promote bounded evidence from success, recovery, and correction only through governed, reversible controls.
AeroClear turns fragmented preflight data into one mission decision surface.
Plan an area of interest, compare hourly launch windows, evaluate aircraft limits, review weather and space-weather context, inspect nearby crewed-aircraft activity, and export a structured flight report from one responsive platform.
AeroClear assists planning. It does not replace pilot judgment, regulatory authorization, or required preflight procedures.
Focused products built from the same SALT19 doctrine.
EvoMind is the research architecture. SALT19’s applied products demonstrate the same engineering discipline in bounded operational domains.
EvoMind
Flagship governed cognitive architecture for persistent reasoning, capability composition, and verified execution.
AeroClear
Responsive mission-planning SaaS for weather, GNSS context, terrain, and exportable flight reports.
SignalLoom
Local-first, human-gated invoice workflow agent for extraction records, AP packets, and routing payloads.
Open SignalLoom →UAS Survey Tool
Windows desktop tool for KMZ AOI loading, GCP layouts, terrain support, and Google Earth deliverables.
Open UAS Survey Tool →Patriot Claim Tracker
Chrome tool for organizing VA.gov claims, messages, documents, and forms into a local-first dashboard.
Open Patriot Claim Tracker →Recruiter.ai
Resume-first job matching assistant for exact listings, freshness signals, and direct apply links.
Open Recruiter.ai →Keep the workspace close to the user.
SALT19 favors device-native and privacy-aware operation where practical so files, workflow context, and memory do not have to be surrendered to a remote service simply to make AI useful.
How SALT19 Works →Verification and control belong inside the architecture.
Reason before action, authority before execution, verification before belief, evidence before capability claims, and human control at consequential boundaries.
Open Validation Record →A public record for the architecture and evidence.
SALT19 separates research hypotheses, published mechanisms, runtime demonstrations, benchmark evidence, and current engineering validation instead of collapsing them into one marketing claim.
Multi-capability composition
Recent bounded validation exercises a high-level objective becoming multiple required work products through semantic capability resolution.
Verified artifacts + lineage
Execution evidence distinguishes attempted operations from identifiable resulting artifacts with version and provenance information.
Governed experience reuse
Repeated success and correction are treated as separate evidence classes for bounded, reversible capability reuse.
AeroClear Operations
Continuing to ship operational software alongside the research program, providing UAS flight intelligence and mission planning.
What SALT19 is—and what it is not claiming.
What is SALT19?
An independent applied AI systems lab researching governed cognitive systems and building practical software for real-world work.
What is EvoMind?
SALT19’s flagship governed cognitive architecture for persistent reasoning, memory, planning, capability composition, and verified execution.
Is EvoMind AGI?
EvoMind is an experimental AGI-oriented cognitive architecture under active validation. SALT19 does not present it as externally certified AGI.
What is AeroClear?
SALT19’s UAS flight-intelligence and preflight mission-planning platform for weather, GNSS context, terrain, and launch-window analysis.
How do you validate AI work?
We separate research hypotheses, architecture claims, runtime demonstrations, and engineering validation on dedicated public pages.
What does local-first mean?
Keeping files, workflow context, and sensitive review state close to the user while retaining explicit human authority at consequential boundaries.
Research-led systems. Human authority.
SALT19 is testing a concrete systems hypothesis: useful autonomy should combine persistent cognition, explicit authority, verifiable outcomes, and controlled learning.