Evidence Desk by LLM Squared

Prepare the file. Keep the open questions.

Run the local preparation methods in this browser. Begin with one declared AI use, or try a review example. Your institution keeps the decision.

No account or assistant installation needed. Inputs remain in page memory. Export before closing the tab.

Choose a review and supply declared metadata.

Exclude member data, account numbers, credentials and confidential documents. The guard screens common restricted patterns; it cannot detect every sensitive detail. Illustrative inputs are not a client record.

Review

AI useName the task, not a customer.Where it runsVendor service, institution system or public assistant.Data categoriesCategories only. Exclude member data, account numbers and credentials.Owner roleA role is enough. Leave unknown fields blank.Evidence on handDescribe the permitted document. No document upload is needed.Supplied referenceA reference label does not verify a source.Question still openName the answer the reviewer still needs.

Structured input and complete supporting records

Input to prepare

Use structured input for multiple uses, supplied tests, quotes, source excerpts or retained change records. Edit the example only with declared or redacted information. Internal reference labels belong in this supporting record.

Declared input JSON

Preparation, expert review and acceptance are separate.

Evidence Desk uses the same 0.2.0 deterministic preparation logic as the local Claude package. It preserves failed cases, missing terms and supplied links; it does not retrieve or judge the sources.

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

Ask LLM Squared to review permitted sources. Use the local Claude package. Read the data-handling rules.

Prepare the file. Keep the open questions.

Choose a review and supply declared metadata.

Declare agent permission.

Prepared file and supporting records.

Preparation, expert review and acceptance are separate.