EvoMind →Agent Utility Grid →AeroClear →Products →DroneMesh →TacWing-X →CloneCode →AI Dispatch →ARCS Research →Updates →Research →About →
Reality-verifying AI · UAS · robotics · local-first systems

The bottleneck is no longer intelligence.
It is trusted execution.

SALT19 builds governed, verifiable AI systems that bridge the gap between model intelligence and real-world execution. The goal is not merely to issue actions, but to preserve authority, observe outcomes, verify state change, and attach evidence to what the system believes happened.

UAS flight intelligenceDrone photogrammetryAutonomous systemsLocal AI engineeringMachine-native servicesGoverned cognition
EvoMind governed cognitive architecture An abstract neural system showing a governed flow from interpretation through composition and execution to verification. Interpret Compose Govern Verify
persistent cognition · capability composition · governed execution · verified outcomes
I am a
From tool use to reality verification

Execution is a claim. Verification is evidence.

Most software agents can issue an action and receive a successful API response. SALT19's systems work toward the harder boundary: separate intent, execution, observation, verification, and belief admission so an autonomous system has evidence for what actually changed.

Claim boundary: this visualization explains a systems principle used across SALT19 research and engineering. It does not claim that every SALT19 product controls physical actuators or that all verification problems are solved.
Tool-using pattern
POST /action
→ 200 OK
“Task complete.”
UNVERIFIED No independent observation has been admitted.

A transport response can prove that a request was accepted. By itself, it does not prove the intended external state now exists.

Reality-verifying pattern · local visualization

Intent → Execute → Observe → Verify → Admit

01Intentdesired state
02Executegoverned action
03Observenew evidence
04Verifycompare claim
05Admitupdate belief
INTENT REGISTEREDNo success credit yet. Evidence is still required.
Why this matters

Deep tech should survive contact with reality.

Three recurring design pressures connect SALT19's work: operational authority, sovereign/local computation, and evidence strong enough to challenge an agent's own claims.

AeroClear UAS flight intelligence interface
01 · Operational authority

Mission software must respect the operator.

AeroClear and TacWing-X are built around operational context, explicit authority boundaries, and human-supervised use rather than treating an AI output as flight authorization.

See the TacWing-X authority model →
DroneMesh locally processed photogrammetry presentation
02 · Sovereign computation

Useful systems should not require permanent cloud custody.

DroneMesh and local-first SALT19 tools keep important processing paths close to the operator, reducing dependence on centralized services when privacy, resilience, or field conditions matter.

Explore DroneMesh →
Closed-loop cognitive system architecture diagram
03 · Evidence over assumption

Agents need a way to be wrong about their own success.

EvoMind's governed cognitive work emphasizes verification, memory, planning, and inspectable execution so progress can depend on evidence rather than fluent self-report.

Open validation evidence →
Proof lab

Use the website, not just the brochure.

These demonstrations run in the browser. The DroneMesh viewer loads static site assets only after you ask for it. The Agent Utility Grid walkthrough is intentionally offline and creates no Worker or D1 traffic.

Interactive geospatial proofDroneMesh orthomosaic viewer
Static assets · click to load
Drag. Tilt. Inspect.

Launch the existing DroneMesh browser viewer and interact with a locally processed orthomosaic presentation. The carved-earth effect is presentation geometry, not measured elevation.

Machine-commerce contract walkthroughTry the Grid without touching the Grid
0 Worker calls · 0 D1 ops
local-simulation://agent-utility-grid
$ discover utility

{
  "tool": "example-go-no-go",
  "transport": "MCP / HTTP",
  "commerce": "x402"
}

# Local demonstration only. Nothing was sent.

This is intentionally not a live dashboard. It is a deterministic client-side walkthrough of the request → payment requirement → settlement authorization → structured result → receipt shape. It does not contact api.salt19.com, broadcast a payment, query D1, or add load to the production Agent Utility Grid.

Validation as a story

Challenge → architecture → evidence.

Technical credibility is easier to understand when the evidence trail is attached to a concrete engineering problem instead of presented as a wall of specifications.

Governed cognitive systems

How do you stop an agent from awarding itself success?

Challenge

Action completion and state verification are different claims.

Architectural approach

Separate planning, governed execution, observation, verification, and durable evidence.

Public evidence

Validation reports and DOI-backed EvoMind research expose the architecture and its tested boundaries.

Review the validation record →
Local geospatial production

Can drone imagery become a useful deliverable without a cloud-first workflow?

Challenge

Photogrammetry pipelines can be costly, opaque, and operationally dependent on remote services.

Architectural approach

DroneMesh is being engineered as a local-first desktop production pipeline with explicit evidence and output surfaces.

Public evidence

The site exposes real processing outputs, visual comparisons, and an interactive orthomosaic presentation.

Open DroneMesh evidence →
Field AI under real constraints

What does a beta look like when the misses are not edited out?

Challenge

Perception demos can look perfect when failure frames disappear from the story.

Architectural approach

TacWing-X keeps operator authority explicit while local perception, tracking, OCR, telemetry, and federation mature.

Public evidence

The lab record shows real model errors alongside working capability rather than presenting beta output as production certification.

Open the TacWing-X lab record →
RADICAL HONESTY

Yes, the beta briefly called a roughly 200 lb Presa Canario a cow and a table a chair. Those frames remain part of the public TacWing-X story because an honest failure boundary is more useful than a perfect montage.

See the unpolished evidence →
Investors & strategic partners · conversations open

The bottleneck is no longer intelligence. It is trusted execution.

SALT19 is building the governed software layer between increasingly capable AI and real-world work: machine-native services, UAS flight intelligence, local geospatial production, supervised autonomy, AI-assisted engineering, and persistent cognitive systems.

SALT19 in one sentence: one reusable engineering spine for governed AI, machine commerce, UAS operations, geospatial production, and autonomous systems — with multiple routes to market and public proof surfaces.

SALT19 is a private company. This page is informational and is not an offer to sell securities.

01 · Why now

AI is moving out of the chat window.

Model capability is accelerating while real deployment still needs governance, verification, provenance, local execution, and domain-specific integration.

02 · Technical edge

A shared systems spine compounds.

Cognition, verification, machine services, UAS intelligence, geospatial processing, and operator-control patterns are engineered as reusable system capabilities rather than isolated demos.

03 · Distribution

More than one path to a buyer.

Commercial paths include software subscriptions, local desktop licensing, enterprise and government deployment, developer infrastructure, and machine-to-machine paid utilities.

04 · Evidence

Proof is inspectable.

Live products, production machine endpoints, public engineering updates, DOI-backed research, and validation records give investors a diligence trail beyond presentation slides.

Publications & research · continuously updated

The systems are shipping. The record is public.

AI Dispatch turns Gabriel Allit's AI and AGI writing into a first-party technology publication. ARCS and SALT19 research surfaces preserve DOI-backed records, research discovery, validation boundaries and engineering updates.

40K+combined social following · Sep 2026
44AI Dispatch + indexed research records
ARCS + Lab record

Research discovery and engineering updates in one public trail.

Explore the ARCS research community, Research Digest and Radar, then follow SALT19 development updates across UAS, agent infrastructure, governed cognition, geospatial production and local AI systems.

September 2026 · Product Update

AeroClear is expanding. DroneMesh, TacWing-X, and CloneCode are next.

AeroClear is live with a broader aviation-data picture through FAA SWIM integration. At the same time, SALT19 is advancing three new systems for drone-data production, UAS operations, and local AI software engineering.

AeroClear UAS flight-intelligence and mission-planning dashboard
AeroClear · live UAS flight intelligence and mission planning
DRONEMESH · LOCAL WORKSTATION
Mission processing
Outputs
TACWING-X · OPERATOR DISPLAY
Platform state
PNT VALID
GNSS + INS
LINK 92%
BATTERY 81%
WORLD TRACKS 04
AUTHORITY OPERATOR
CLONECODE · LOCAL AI WORKSPACE
src/
 core/
 ui/
tests/
README.md
pyproject.toml
def verify(change):
  tests = run_tests()
  if tests.green:
    return review(change)

# local model route
mode = "Agent"
LOCAL AI

Plan prepared.
RED evidence captured.
Implementation bounded.
Fresh verification pending.
Live nowAeroClear + FAA SWIMFlight intelligence, traffic context, mission planning and reporting.
Coming soonDroneMeshLocal-first photogrammetry and geospatial production.
In developmentTacWing-XUAS operator display, telemetry and supervised autonomy.
Active validationCloneCodeLocal-first AI coding workspace for Windows.
DroneMesh · Geospatial Production

A real drone mission becomes an interactive site.

DroneMesh turns aerial imagery into immersive mapping products from a local-first desktop workflow. Explore the first orthomosaic, interact with the carved-earth presentation, and see where SALT19 is taking drone-data delivery.

Local-firstOrthomosaicsElevationPoint clouds3D delivery
DJI Mavic 3 · 400 ft AGL DroneMesh carved-earth orthomosaic of a solar site DroneMesh · First OrthomosaicCaptured by drone · locally processed · digital twin ready
SALT19 Systems

Live products now. New systems coming next.

From UAS flight intelligence and machine-native services to drone-data production, autonomous operations and local AI engineering.

Available now

Systems you can explore today

Operational software, production infrastructure, and public research surfaces.

Live product

AeroClear

UAS flight intelligence for weather, GNSS/Kp, terrain, nearby cooperative traffic, launch windows, FAA SWIM aviation data and flight reports.

Explore AeroClear →
Production infrastructure

Agent Utility Grid

Machine-native technical services exposed through MCP with x402 payment gating, structured outputs and verification evidence.

Explore Agent Utility Grid →
Research + validation

EvoMind

Persistent reasoning, memory, planning, governed execution, verification and controlled experience-driven improvement.

Explore EvoMind →
Coming next

Three new SALT19 systems in active development

Public previews describe what exists now and clearly separate current capability from future scope.

DroneMesh running on a SALT19 workstation REAL DRONEMESH INTERFACE
Coming soon

DroneMesh

A private, local-first photogrammetry workstation that turns drone imagery into orthomosaics, elevation products, point clouds, 3D geometry and field-ready export packages.

3,711DJI M3E images
18m 22send to end
10validated deliverables

Completed full-mission production run on a local workstation. Runtime is specific to this mission, profile and hardware.

Preview DroneMesh →
TACWING-X · OPERATOR DISPLAY
Platform state
PNT VALID
GNSS + INS
LINK 92%
BATTERY 81%
WORLD TRACKS 04
AUTHORITY OPERATOR
In development

TacWing-X

An Android operator display and edge-autonomy foundation for live telemetry, resilient localization, mapping, platform state and human-supervised robotic operations.

Preview TacWing-X →
CLONECODE · LOCAL AI WORKSPACE
src/
 core/
 ui/
tests/
README.md
pyproject.toml
def verify(change):
  tests = run_tests()
  if tests.green:
    return review(change)

# local model route
mode = "Agent"
LOCAL AI

Plan prepared.
RED evidence captured.
Implementation bounded.
Fresh verification pending.
Active validation

CloneCode

A Windows-first, local-first AI coding workspace with local models, repository intelligence, editor/LSP tooling, Git review and governed agent workflows.

Preview CloneCode →
5DOI-backed EvoMind public research records
9 / 9targeted canonical-path acceptance cases passed
252 / 252surrounding regression checks passed
3 work productsWord, Excel and PowerPoint from one bounded high-level objective
Current evidence: 5 DOI-backed records · 9 / 9 targeted acceptance cases · 252 / 252 surrounding regression checks · 3 verified work products. Internal software validation · August 2026 · architecture, execution, and artifact evidence.
Flagship Research System

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

Consequential execution requires explicit admission and control boundaries, keeping planning, authority, action, and verification distinct.

Production · MCP Registry 0.5.1

Agent Utility Grid gives software agents a commercial execution surface.

Compatible agents can discover SALT19 technical services through the official MCP Registry and machine-readable endpoints, receive a real x402 payment challenge, settle in Base USDC, execute the canonical paid tool, and receive structured results with verification and reconciliation evidence.

Official MCP Registry Native MCP tools x402 v2 Base mainnet USDC Decision receipts

MCP Registry Published

io.github.GabrielAllit1/
salt19-agent-utility-grid

Version 0.5.1 points compatible clients to https://api.salt19.com/mcp.

Live paid utilities

Technical preflight, reliability, and bounded decision utilities are live behind the canonical x402 payment authority.

Current implemented products

Demand-gated remaining demand-gated candidates exist in source with schemas, bounded prices, and claim boundaries before promotion.

Phase 1 Reliability Suite Live

Five Phase 1 products are active, including code-change risk gating, plan verification, claim/output verification, repository adoption, and dependency-upgrade decisions.

Commercial activity is admitted to the revenue record only after payment settlement and successful execution are verified against the same transaction.

Cognitive Architecture

A control architecture for cognition & evidence.

Organized around persistent state, reasoning and planning, semantic capability resolution, governed admission, real execution, and outcome verification.

01Human goal
02Persistent cognition
03Capability plan
04Resolution
05Governed admission
06Real execution
07Verified outcome

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.

UAS Flight Intelligence

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 cooperative crewed-aircraft activity, incorporate aviation information delivered through FAA SWIM data services, and export a structured flight report from one responsive platform.

Live launch-window scoring GNSS & Kp context Crewed-aircraft awareness 3D terrain & AOI
AeroClear responsive UAS mission-planning dashboard

AeroClear is designed to support pilot-in-command decision-making alongside regulatory authorization and required preflight procedures.

Local-First Doctrine

Keep the workspace close to the user.

SALT19 favors device-native and privacy-aware operation where practical so files, workflow context, and memory can remain close to the user.

How SALT19 Works →
Bounded Authority

Verification and control belong inside the architecture.

Reason before action, authority before execution, verification before belief, evidence before capability claims, and explicit policy around consequential autonomous behavior—including machine-to-machine commercial decisions.

Open Validation Record →
Research + Validation

A public record for the architecture and evidence.

SALT19 separates research hypotheses, published mechanisms, runtime demonstrations, benchmark evidence, commercial telemetry, and current engineering validation instead of collapsing them into one marketing claim.

Multi-capability composition
Verified artifacts + lineage
Agent Utility Grid 0.5.1
AeroClear Operations

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.

Agent Utility Grid

Production machine infrastructure adds externally discoverable MCP distribution, canonical x402 payment gating, and bounded commercial telemetry to the SALT19 systems portfolio.

AeroClear Operations

Continuing to ship operational software alongside the research program, providing UAS flight intelligence and mission planning.

FAQ

What SALT19 is building.

What is SALT19?

An independent applied AI and autonomous-systems lab building UAS flight intelligence, drone-data production, autonomous operational software, local AI engineering tools, machine-native services, and governed cognitive systems.

What is EvoMind?

SALT19’s flagship governed cognitive architecture for persistent reasoning, memory, planning, capability composition, and verified execution.

What is Agent Utility Grid?

Production machine-to-machine infrastructure for AI agents to discover and invoke technical utilities through native MCP and machine-readable surfaces, with canonical x402 payment gating and bounded decision receipts.

Is it in the official MCP Registry?

Yes. Version 0.5.1 is published as io.github.GabrielAllit1/salt19-agent-utility-grid with remote MCP endpoint https://api.salt19.com/mcp.

What is EvoMind’s AGI direction?

EvoMind is SALT19’s AGI-oriented governed cognitive architecture focused on persistent reasoning, memory, planning, capability composition, desktop execution, verification, artifact provenance, and controlled experience-driven improvement.

How does Agent Utility Grid verify commerce?

The Grid verifies payment settlement and successful execution against the same transaction before admitting commercial activity to its auditable revenue record.

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, engineering validation, commercial telemetry, and release evidence on dedicated public surfaces.

Research-led systems. Bounded real-world execution.

SALT19 is testing a concrete systems hypothesis: useful autonomy should combine persistent cognition, explicit authority, verifiable outcomes, controlled learning, and—where software acts economically—bounded machine-to-machine commercial authority.

Open Research Community

ARCS — Autonomous Reasoning, Cognitive Systems & Agentic AI

A public research community hosted by SALT19 for papers, preprints, software, datasets, benchmarks, replication work, protocols, and documented negative results across reasoning-centered and increasingly general agent systems.

Community infrastructure

Zenodo-backed records. Automated public discovery.

Zenodo remains the DOI-backed archive and curation authority. SALT19 automatically materializes current records, an RSS/JSON feed, research digests, record spotlights, and a separately labeled external research radar for public discovery.

Federal vendor identity

SALT19 LLC is active in SAM.gov.

SALT19 maintains an active U.S. federal entity registration for award activity. The identifiers below are published so acquisition teams, prime contractors, partners and customers can verify the business quickly without relying on marketing claims.

SAM.gov registration
Active
Annual renewal date: Aug 31, 2027
Unique Entity ID
TXCRKHSCFN94
SAM.gov entity identifier
CAGE Code
18DJ8
Defense Logistics Agency CAGE Program
D-U-N-S®
11-758-0128
Dun & Bradstreet business identifier

Registration status and identifiers establish business identity and federal registration. They do not imply a federal award, contract, certification or government endorsement.