Local-first AI and digital attribution

Software becomes bespoke. Trust cannot.

SwarmTracker is building local-first AI infrastructure to investigate digital evidence, trace it to its source where the evidence supports it, and verify what AI systems were authorized to do and what actually changed.

Introducing SwarmTracker

Our original introduction. An updated overview is coming soon. Watch on Vimeo

Local-first by design

SwarmTracker Node is an appliance in development designed to keep analysis, policy enforcement, and decision records on site. It brings together Verity Local for analysis, Keel for policy and authorization, the model runtime, and a local ledger. Verity Cloud separately supports reports and accounts. Onboard vessel use is one application of this local-first design.

From evidence to verified outcomes

  1. 01 / Sources
  2. 02 / Findings
  3. 03 / Decisions
  4. 04 / Outcomes

Investigate digital evidence.

Cyber attribution starts with preserving where evidence came from and examining what it supports about the source of an activity. Contradictions and uncertainty stay visible. Try the public tools.

Govern and verify AI actions.

Record what an agent proposed, who approved the exact action, what the executor did, and what a separate check confirmed. Those records let someone reconstruct the decision and its outcome later.

Solutions

Verity

Verity’s public analytical interface gives people a way to inspect the evidence behind a finding. We’re developing additional workflows for reviewing proposed AI actions and verifying their results. Keel handles policy and authorization in those workflows.

Explore Verity

Galene™· Marine product in development

Galene applies the same foundation to vessel documentation, maintenance records, and crew handovers. Its initial focus is helping crews find relevant information and carry an accurate record from one watch to the next. Crew members retain authority over decisions and actions.

Explore Galene

Controlled Galene pilot using synthetic records

What a verified record update looks like

An agent proposed an update to a maintenance handover. Keel recorded a person’s approval of the exact change, and a protected executor recorded one attempt. A separate verifier read back the saved entry and confirmed what was written:

Handover record
replacement delivered, installation unconfirmed.

The next crew would see that the replacement was recorded as delivered and that installation still needed confirmation.

Technical notes
  • Browser approval: passed in the controlled pilot.
  • Independent read-back: passed in the controlled pilot.

Threat Intel Brief

A curated stream of cybersecurity updates from trusted sources.

View all Signals & Insights

Public Tools

  • PhishFinder™ helps assess suspicious links, domains, and messages.

    Open PhishFinder
  • Deepfake Defender helps examine potentially manipulated media and shows the limits of the assessment.

    Open Deepfake Defender

Research

Our research includes Origin and physics-constrained attribution, which examines whether evidence fits physical limits. We’re also investigating inference-time governance, epistemic validation to assess how well evidence supports a claim, and resilient information sharing when some participants or sources may be unreliable.

Read the Framework Paper

Intellectual property

Patent applications

  • Constitutional AI GovernanceU.S. application: Notice of Allowance receivedPCT application filed

    Turns governance rules into controls that an AI model must follow while it generates output, and keeps a tamper-evident record of what was enforced.

  • Physics-Constrained AttributionU.S. application pendingPCT application filed

    Tests whether the evidence behind a cyber attribution is consistent with physical constraints and independent evidence.

  • Epistemic Validation / Byzantine-Resilient ExchangeU.S. application pending

    Assesses the quality of evidence, re-validates intelligence after privacy-preserving transformations, and exchanges it in a way designed to withstand faulty or adversarial participants.

Trademarks

  • SWARMTRACKER®U.S. Principal Register · Reg. No. 8419041 · Registered Official record
  • DEEPFAKEDEFENDER®U.S. Principal Register · Reg. No. 8488956 · Registered Registered to Kevin McGuinnessOfficial record
  • PHISHFINDER®U.S. Supplemental Register · Reg. No. 8267242 · Registered Registered to Kevin McGuinnessOfficial record

Talk to us

If you’re investigating digital evidence, evaluating governed AI, or exploring onboard assistance, we’d like to hear what your workflow requires.

Kevin McGuinness, Founder · kevin@swarmtracker.ai