Skip to main content
LLM Squared

Independent of Jack Henry.

Legal

Terms of Use

Last updated October 3, 2026 LLM Squared info@llmsquared.com

Decision tool results

Policy text matches do not establish an answer, compliance, policy adequacy or enforce access rights. Workflow value estimates are modeled planning results, not achieved savings or investment approval. The responsible owner must review inputs, sources and interpretation.

Assistant downloads are provided for evaluation through a compatible client. Directory acceptance and client installation are separate. The packages do not authorize transactions, account access or document transfer to LLM Squared.

Invoice Evidence preview

Invoice Evidence compares entered structured fixed-rate USD data. Source confirmation, permission and completeness are reviewer declarations. Modeled discrepancies do not establish contractual rights, refunds, achieved savings or approval to pay. Unsupported adjustments and missing evidence require a different review. Quantity limits apply only within the supplied invoice packet.

Vendor evidence and declared permissions

Vendor evidence mappings and evidence type matches do not establish document sufficiency, compliance, contractual rights, risk acceptance or vendor approval. Agent Permission Review does not prove identity, effective permissions or enforcement. A complete declaration set does not grant action permission. Review Workspace hash parity does not authenticate a source or reviewer. Human owners must inspect the actual evidence and decide within their authority.

Connected case results

Case orchestration retains each workflow’s own sources, dates and qualification. It does not resolve conflicts between workflows or certify the source material. Completed means a report was prepared, including reports with unresolved evidence or blocked source use. Input and output hashes bind matching JSON values; they are not signatures. A human review note remains a declaration. Export checks recompute the current engines and do not authorize a business action.

Exact declared case changes

Review Comparison requires an exact matching case ID and matches supplied job IDs. It shows declared value, label, workflow and ordered job-list changes. Both versions rerun through the current engines. A changed execution status or report does not establish corrected evidence, source rights, risk reduction, approval, refunds or achieved savings. Missing and present null values remain distinct. Optional reviewer details are supplied declarations. Current export parity binds the generated comparison to the supplied JSON values.

Policy library evidence

Source rights, approvals and controlling versions are supplied declarations. Any blocking source issue suppresses the entire collection's excerpts. Literal phrases require human interpretation and do not establish policy adequacy or applicability. Content hashes bind exact supplied UTF-8 text, not original document authenticity. Follow-ups reflect recorded issues and literal coverage; they do not assign tasks, verify risk or infer deadlines. Export parity is freshly checked.

Pilot comparisons

Eligible comparisons use only complete matched supplied observations and declared quality, permissions and reviewer acceptance. Declarations and references are not authenticated. Labor-cost differences use supplied rates and do not prove payroll reduction or cash savings. Excluded samples remain visible; a sample does not establish causality or organization-wide impact. Export parity is freshly checked.

Portfolio, change and access preparation

PowerOn Portfolio Studio and Change Review provide literal, heuristic inventory and candidate-path preparation from supplied text. They do not compile, execute, establish complete runtime dependencies, repair or deploy code. Source rights, ownership, environment and plans remain unverified declarations. Agent Access Lab provides deterministic simulation of supplied scopes and events; it does not verify identity, register an agent, grant authority or transact.

Studio, decision brief and portable packages

Product Studio combines supplied-input preparations without establishing causality between them. Pilot Decision Room records scope, proposed acceptance and optional cost hypotheses; its output is not an accepted offer, agreement, guaranteed outcome or achieved savings. A scoped engagement requires separately agreed fee, deliverables, rights, handling, reviewer responsibilities, acceptance and support. Download availability does not establish client installation, model activation or public directory acceptance.

Use only inputs you are authorized to process, and keep full-source exports within approved storage and sharing. A qualified institution reviewer must verify actual runtime or production decisions independently.

Knowledge Acceptance Lab

The lab checks supplied questions, answers, source excerpts, citation/version records and human review declarations in browser memory. It does not generate answers or verify their meaning, reviewer identity, permissions or institutional approval. Context fingerprints help flag changed review records; they do not prove authenticity. Full JSON and printable outputs include every supplied input. Use approved storage and sharing, and save work before closing. No source URL is fetched and no browser storage is used.

Pilot Evidence Preparation

This local browser method keeps supplied evaluation CSV, required cases and explicit row declarations in browser memory. Nothing is submitted; there is no account, cloud history, monitoring or reminders. Downloads are files you choose to keep. Provider results do not establish institution acceptance, source permission, identity or actual testing. Local packages do not activate a model or establish directory acceptance. No tenant/core connection, financial action, compliance certification, legal opinion or achieved savings is provided.

Requests and device notes.

The contact form creates a scope inquiry. It does not reserve an appointment, accept an engagement, authorize access or subscribe you to marketing. You choose whether to send prepared notes. Review the current text before transmission. The institution approves its source-handling route before any review begins.

A quarterly refresh of the inventory file requires a separate agreed cadence, source boundary, fee and review responsibility. The current local method follows declared relationships; it is not continuous monitoring.

Evidence Desk use.

Supply only declared or redacted inputs you are allowed to share. Exclude member data, account numbers and credentials. The input screen is an additional safeguard, rather than approval to submit restricted information.

Outputs prepare questions and evidence from supplied sources. The institution's reviewer owns controlling-version, contract, acceptance and release decisions. Failed, missing and unknown records remain visible.

A review prepares a file. It does not certify compliance, give a legal opinion or connect to a core.

Read the current server route and limits. Support: info@llmsquared.com.

Local assistant evaluation package.

The Claude Code package is an evaluation download. Installation does not activate a provider account, create an engagement or establish a public directory listing. Use an institution-approved environment and permitted declared inputs.

The optional export helper preserves supplied records and a blank decision; the institution reviewer remains responsible for acceptance. Third-party components keep their license files. This download does not add an open-source license to LLM Squared code.

A review prepares a file. It does not certify compliance, give a legal opinion or connect to a core.