Welcome to our products page
UVLM develops local-first tools for provenance, evidence review, artifact integrity, and human-governed artificial intelligence. Every offering carries a visible release status, a bounded use case, known limitations, and a clear access path.
Open-Source Alpha Local First AI Modules, Contract, and Engine.
UVLM products
Public open-source research preview
UVLM IntegrityLock Community Edition
Hash exact local files, build closed package records, create deterministic archives, verify manifests and signatures, preserve provenance events, compare release lineages, and create optional encrypted package vaults—without an account, activation, payment, model provider, or remote service.
Public open-source research preview
UVLM Critical Epistemic Continuity Protocol Community Edition
Evaluate structured action-and-evidence proposals against the fixed UVLM-CECP 1.1.0 contract and generate deterministic PASS, HOLD, or REJECT receipts—without an account, activation, payment, model provider, network service, persistent memory, or training.
Public open-source research preview
ΔSyn Conditional Compression Lab
Test whether an explicitly chosen donor reduces complete description length after counting reconstruction, prior synchronization, payload, residual, provenance, and deterministic gzip.
Public open-source research preview candidate
Governed AHA Lab
Turn a cross-domain analogy into a source-grounded structural bridge, explicit disanalogies, a target prediction, and the smallest falsifying test—without calling a model or selecting donors for you.
Public open-source research preview
Echo Writing Primer Community Edition
An open, local editorial-governance toolkit for preserving source fidelity, active agency, distinctive voice, role separation, and human final authority.
Public open-source research preview · preconformance
Trauma Informed AI Protocol (TIAIP) for CoherenceLattice
Review AI-generated candidates through a thin, stateless governance sidecar that preserves voice, uncertainty, choice, source boundaries, and bounded repair without diagnosing trauma.
Public open-source research alpha · Local-first web stewardship
Research Alpha — Human Review Required
UVLM Open Web Steward Community Edition
Turn controlled product facts into one bounded landing-page candidate, then inspect, repair, and export the evidence without giving a model publication authority.
The single-file local workbench creates Micro or Standard task capsules for a separately chosen model, quarantines returned HTML as an untrusted candidate, builds an inert preview, runs deterministic web checks, preserves skipped lanes, and requires an explicit human decision.
- v0.1.0-alpha.2
- MPL-2.0 + CC BY 4.0
- No account or activation
- No bundled model
- No application network calls
Controlled facts → candidate → human decision
Open Web Steward standardizes a bounded web-production lifecycle without pretending that model fluency, a passing check, or a checksum is the same as truth, accessibility conformance, security, or permission to publish.
- Controlled facts
- Teaching capsule
- Untrusted HTML
- Native findings
- Human decision
- Bounded repair
- Evidence export
Use boundary. This is a research Alpha for one bounded product landing page. Production use is not authorized, and human content, accessibility, security, legal, CMS, and live-site review remain required.
Public open-source alpha · Local runtime · No charge
UVLM CoherenceLattice software
432 Atlas + GUFT Rosetta Bridge
UVLM CoherenceLattice Optional Atlas and Rosetta Modules
A bounded software implementation of the UVLM 432 Universal Pattern Bridge: a source-grounded 432-archetype reference Atlas plus Waveform and Recursive Geometric Fiber Rosetta methods, joined through the Grand Unified Field Theory of Coherence (GUFT) / Universal Coherence Mapping (UCM) vocabulary and the exact CoherenceLattice alpha.3 validation boundary.
Contextual humanistic addresses—not person types
Each record provides an address i, a local expression coordinate s in [0,1], source and review posture, and permission to return competing readings or no adequate mapping. The full local identity is (i, s), never a free-floating score of a person.
Expression, form, residual, and refusal remain distinct
The Waveform Rosetta reflects expressive residual around a cell master; the Geometric Fiber Rosetta opens a typed local space inside the address. The packet keeps GUFT axes, geometry, uncertainty, provenance, consent, and off-axis residual visible rather than forcing unlike domains into one substance.
- Release v1.3.0
- 432 records
- Waveform Rosetta v2
- Geometric Fiber v1
- MPL-2.0
- Not production ready
Common grammar, not common substance. GUFT coordinates identity, state, relation, uncertainty, residual, provenance, and refusal while native units, causal theories, and evidence standards remain sovereign. This alpha does not classify people, certify science, or create authority.
Public open-source alpha · Developer contract
Local-first · Model-agnostic · Authority effect: noneCoherenceLattice Product Module Contract
Make products speak CoherenceLattice without rebuilding the cognition engine around every product.
A small executable contract for governed requests, quarantined candidates, typed findings, bounded repair authority, human decisions, telemetry, and replay receipts. Native product logic remains native; the contract standardizes the anatomy, not the personality.
- v0.1.0-alpha.6
- MPL-2.0
- Python 3.11+
- 9 contract objects
- 10 module roles
- No model required
One governed lifecycle
Request → candidate → findings → human repair decision → bounded successor → revalidation → human disposition → replay receipt.
Products keep their expertise
Each module retains its own validators, protected surfaces, evidence rules, and product-specific meaning.
Caller paths are not disposable
The alpha.6 successor removes recursive output deletion and rejects unowned, altered, linked, or unknown output content.
Research boundary. Contract conformance is not truth, product approval, accessibility or security certification, legal or clinical authority, mapped-product release, or permission to merge the interface into the CoherenceLattice core.
Public open-source research preview
CoherenceLattice Operational Cognition Engine
A local-first, provider-agnostic Python research engine for authenticated source bundles, typed candidate observations, bounded structural mapping, explicit uncertainty, deterministic receipts, package verification, and exact replay.
- v0.1.0-alpha.3
- MPL-2.0
- Python 3.11+
- No account
- No model required
- No network required
Boundary: A hash establishes byte identity, not truth. A receipt records a process posture, not a diagnosis, person classification, compliance certificate, or final answer.
How UVLM Tools Can Work with Local AI
UVLM software may be used beside a locally operated AI model or another user-selected tool without requiring source material to be sent to UVLM. The local model remains a separate component selected and operated by the user. Each UVLM release states which adapters, interfaces, operating systems, and workflows were actually tested. A local model may generate a candidate response, classification, summary, or artifact. UVLM software can help ground, quarantine, review, document, compare, verify, or seal that candidate according to the function of the particular release. Model output does not become true, approved, safe, lawful, or authorized merely because it is fluent or because it passes a technical check. UVLM releases do not silently train a model, change model weights, write persistent memory, activate a remote service, or send customer files off the device unless the exact release documentation expressly states otherwise. Users remain responsible for the model, source material, permissions, configuration, outputs, backups, and decisions made from those outputs. Researchers and enthusiasts may compare an ordinary local-model run with a UVLM-governed run. Any claim that a UVLM tool improved performance must identify the exact model, version, quantization, adapter, prompt, source set, hardware, settings, baseline, metric, and uncertainty. Improvements in provenance, review burden, source fidelity, reproducibility, privacy, or correction must not be misrepresented as an improvement in the model’s underlying weights or general intelligence. Read the version-bound Technical Manual, Known Limitations, Release Notes, licenses, and verification instructions before use.
Free Internal Control Authorship for AI and Organizational Deployment.
UVLM Public Research Tool Open-Source Alpha v0.2.0-alpha.4
Global Philosophy Governance Atlas
Plural reasons. Visible disagreements. Human authority preserved.
Explore philosophical, cultural, rights, and governance lenses while preserving internal disagreement, translation loss, cultural jurisdiction, countervoices, and affected-party authority. The atlas includes an offline browser, machine-readable records, validation tools, and a Python package for research and public-interest design.
Recommended first step: Open the product page, review the alpha status and use boundary, then choose the complete source-and-offline-app ZIP or the optional Python wheel. No checkout, account, activation, or donation is required.
- 40 philosophical and cultural lenses
- 10 governance problems
- 11 counterpoint packets
- 170 countervoices
- Offline and local-first
- MPL-2.0 code
- CC BY-SA 4.0 UVLM content
Direct public downloads are available on the product page. No checkout, account, activation, or donation is required.
Use boundary: This is an open research and design tool. It is not a culture or identity classifier, a truth-voting engine, philosophical consensus, source-community endorsement, legal advice, or permission to operationalize sacred, restricted, collective, or community-governed knowledge.
UVLM Public Toolkit UICAF v0.1 July 2026
Union Controls Toolkit
Trust needs a trail.
A practical, bilingual implementation library for labor organizations that want to move from promises to procedures, from procedures to evidence, and from evidence to accountable correction. Each bundle includes a quick-start guide, full professional handbook, ten-sheet planning workbook, ready-to-use tables, optional machine-readable data templates, diagrams, and verification records.
Recommended first step: Read the Quick Start Guide, preserve the original ZIP, and make a working copy of the planning workbook. Tailor that copy to your governing documents, roles, systems, risks, and legal requirements. Do not attempt to implement all 379 proposed controls at once.
- Free public ZIP bundles
- English and neutral Spanish
- 379 proposed controls
- 42 illustrative risks
- 36 implementation actions
- 9 optional data templates
- 6 diagrams
ZIP links open or download directly. No checkout, account, or donation is required. The companion article opens in a new tab.
Use boundary: This is a standards-informed, prospective design and planning toolkit. It is not legal advice, an audit opinion, a compliance certification, proof that a control operated, or a finding about a person or past event. Applicable law and governing documents control.
UVLM E-book · Public Edition v1.0.1
The Cost of Keeping One’s Name
Honor, Meaning, and Survival Inside Dishonorable Systems
What does it cost to remain honorable when institutions reward omission, invert reputations, and turn conscience into infrastructure? This long-form essay follows that burden through lived experience, philosophy, global traditions, and a practical appendix of internal controls.
- Free public PDF
- 96 pages
- Approximately 4 hours
- Published August 9, 2026
PDF opens in a new tab. No checkout, account, or donation is required.
What We Offer
-
How UVLM Releases Software: Bounded, Verifiable, and Honest
No software is perfect. UVLM does not wait for an imaginary state in which every possible defect has disappeared. We also do not release software simply because it runs once, passes a local test, or looks finished.
Our standard is more practical:
A UVLM product may be released when the exact downloadable artifact reliably keeps a clearly bounded promise, fails safely, explains its limitations, and gives the customer a way to verify what they received.
Every product has a release class
Each UVLM download carries a visible release classification. The classification tells customers what kind of use the product is ready to support.
Research Release
Methods, papers, schemas, examples, or research code. These materials support study and experimentation but do not make an operational product promise.Private Evaluation
A working candidate provided to named evaluators in a controlled environment. It may have known limitations and may require direct UVLM support. It is not intended for production work.Bounded Pilot
A small number of identified organizations may use the product within a defined purpose, platform, data boundary, and support arrangement.Public Beta
The product is available more broadly, but some features, accessibility work, integrations, or support processes may still be developing.Production Release
The product is supported for ordinary use within its declared scope. Production release does not automatically mean regulatory certification, clinical validation, legal compliance, or suitability for every high-stakes use.A confidential red-team candidate may also exist before these stages, but it is not a customer release.
What customers receive
A normal UVLM release package includes:
the exact product archive;
a detached SHA-256 checksum;
the product name and version;
installation and verification instructions;
release notes;
known limitations;
licensing and third-party notices;
the supported platforms and connectors;
the product’s privacy, network, memory, and authority boundaries;
a support or issue-reporting route.
Where appropriate, we also provide a public assurance summary describing the testing that was actually completed.
What the checksum proves
The SHA-256 checksum lets a customer confirm that the downloaded file is byte-for-byte identical to the file UVLM released.
A matching checksum can establish file identity and detect alteration. It does not prove that the contents are true, correct, safe for every purpose, legally authorized, or free of every possible defect. That distinction is built directly into UVLM’s IntegrityLock design.
Customers should compare the SHA-256 value of their downloaded file with the value published by UVLM before installing or opening the package.
How we decide that a product is ready
Before release, UVLM evaluates nine areas:
Artifact identity
The version, files, checksums, license, notices, and package contents must agree.Essential user journey
The intended user must be able to install or open the product, complete its main task, understand the result, and save or export the outcome.Safe failure behavior
Invalid input, missing dependencies, timeouts, unsupported models, failed licenses, or malformed output must produce a clear failure or hold. They may not silently become approval.Security and privacy
Enabled features must not expose credentials, ship production private keys, silently activate network services, overwrite customer files, or create hidden memory or training use.Executed testing
Required tests must actually run. A test that was skipped, defined but never executed, or reported only in a builder summary is not treated as passing evidence.Installation and recovery
Customers must have a documented route for installation, verification, updating, rollback, recovery, and removal.Usability and accessibility
The required level depends on the release class. A guided private evaluation may be more technical than a public beta or production product.Rights and claims
The license, third-party obligations, product claims, and supporting evidence must be consistent.Human release approval
A named UVLM human approves the exact artifact, release class, known limitations, and public wording.
Known limitations do not automatically prevent release
A product may be useful and responsibly releasable while still having documented limitations.
A limitation may remain when it:
does not compromise the product’s central safety or data boundary;
does not contradict the advertised function;
is clearly disclosed;
has a containment or workaround;
has an assigned repair path;
does not create an undisclosed high-risk condition.
Critical or high-severity defects on an enabled customer path prevent a wider release. They must be repaired, or the affected feature must be disabled and made unreachable.
Medium and low findings may remain when their practical impact is bounded and honestly disclosed.
Candidate output remains candidate output
Some UVLM products help govern AI-generated material. In those products, model output does not automatically become an approved answer, report, decision, or publication.
For example, the TIAIP organizational workflow preserves model output as a candidate, applies the selected governance mode, records review evidence, and requires a human decision before an official governed export becomes available.
A polished model response is not treated as truth or permission merely because it sounds confident.
Updates preserve lineage
When UVLM repairs or improves a module, the successor receives a new version and a new checksum.
We do not silently replace an earlier artifact while keeping the same identity. Release notes explain what changed, what remained compatible, and whether customers should update immediately.
Older releases remain identifiable by their original:
product name
version
filename
SHA-256
release class
known limitations
Scope matters
Every UVLM product page should state:
who the product is for;
what it is designed to do;
what it is not designed to do;
which platforms and models are supported;
whether it uses local or remote services;
whether it retains memory;
whether customer data leaves the device;
whether the product is an evaluation, pilot, beta, or production release.
A product that implements a framework such as COSO, NIST AI RMF, the EU AI Act, PRISMA, or another standard helps structure work against that framework. It does not automatically certify compliance, replace professional judgment, or create legal authority.
Our customer promise
UVLM does not promise perfection.
We promise to release products that are:
bounded, because the supported use is clearly stated;
honest, because limitations remain visible;
verifiable, because customers can authenticate the exact artifact;
useful, because the central user journey must work;
repairable, because updates preserve lineage and correction history;
human-governed, because consequential decisions do not silently transfer to a model.
A UVLM release is ready when it does what we say it does, on the systems we say it supports, without a known uncontained risk that contradicts that promise.