User Guide

RangerIO is a desktop data-intelligence and sanitization tool for working with files that may contain sensitive information. It imports supported files, builds a local structured profile called a dossier, scans for potential PII, provides review and correction tools, supports local analytical questions, and prepares sanitized or tokenized context for explicitly enabled AI workflows.

In everyday terms, RangerIO helps you open a work file, understand what is in it, check it for sensitive information, ask questions about it, and prepare an output for review or sharing.

What do you need to do?

A good first workflow

RangerIO is not only a chat interface. The data engine operates before a model is involved:

1. 1A source is imported.
2. 2RangerIO examines its structure.
3. 3Local profiling and indexing build reusable intelligence.
4. 4PII scanning records findings and coverage.
5. 5The user reviews the source and scan state.
6. 6The user can query, report, mask, or export the source.
7. 7A local, cloud, or MCP model path is used only when the selected task requires it and that path is available.

RangerIO and RangerIO Plus

In one sentence: RangerIO provides the complete data engine and lets you connect an AI provider. RangerIO Plus adds RangerIO's bundled local-model runtime. Unless a section says otherwise, the data, privacy, dossier, report, and export workflows apply to both products.

RangerIO Enterprise

RangerIO Enterprise is a licensing tier, not a separate product. It runs the same data engine, privacy workflows, and export paths described in this guide, and adds two things for organizations: centralized management for multi-seat deployments, and fully supported database and REST API data sources alongside file imports.

• Centralized management for deployment, configuration, and policy across many workstations.
• SQL database sources as a supported, fully backed import path.
• REST API sources as a supported, fully backed import path.
• Full support for the organizations running these connected sources.

What you can do in both products

• Create projects and project groups.
• Import individual files, multiple files, or selected files from a directory.
• Inspect import tasks and recover an interrupted import.
• View source structure, preview data, and inspect dossier intelligence.
• Review PII findings in tables and documents.
• Mark missed sensitive values, dismiss false positives, reclassify findings, and re-scan.
• Ask questions against one or more ready sources.
• Review grounding, confidence, coverage, and data-quality caveats when supplied with an answer.
• Export data, reports, and supported masked document copies.
• Enable the local MCP server and configure the current Claude Desktop connector.

Where the products differ

RangerIO requires a configured external local or cloud model for model-assisted work. Local dossier lookups and local analytical routes can still operate without a bundled LLM.

RangerIO Plus includes RangerIO's local llama.cpp/GGUF runtime and RangerIO-managed model setup. Selecting that local runtime keeps model inference on the workstation. Cloud and MCP paths remain optional in RangerIO Plus.

Before relying on an output

A completed import, a “Ready” status, a PII scan, a masked export, or an AI answer is not a certification. Review consequential outputs against the source material and your organization's policy.

Terms you will see

Professional work often starts with files that should not be pasted directly into a general-purpose AI chat: contracts, employee files, accounting workbooks, research exports, operational records, and other client material.

RangerIO separates preparing and understanding the source from choosing an AI engine.

The practical problem

Without a preparation layer, a user has to decide between:

• doing the work manually;
• sharing more source content than intended;
• manually removing identifiers;
• rebuilding the same context in every new AI conversation; or
• using an AI answer without a reusable, inspectable source profile.

RangerIO addresses this by building reusable local intelligence first.

The operating model

Local preparation

The data engine imports, profiles, indexes, scans, and stores working intelligence locally.

Explicit model choice

The user can use:

• local dossier lookup or analytics;
• a configured external local provider in RangerIO or RangerIO Plus;
• the bundled local runtime in RangerIO Plus;
• an explicitly configured in-app cloud provider; or
• an explicitly enabled MCP connector.

Review before disclosure

The user reviews scan coverage, findings, masking, and the organizational rules that apply to the data.

The product boundary

RangerIO reduces preparation work and provides safety controls. It does not make an otherwise prohibited disclosure permitted, replace legal review, or guarantee that every sensitive value will be detected.

End-to-end workflow

1. 1Select a source
2. 2Inspect structure
3. 3Import, profile, and build the dossier
4. 4Complete and review privacy coverage
5. 5Analyze, report, or mask
6. 6Export locally or explicitly use an external AI path

Step 1 — Select and inspect

For file imports, RangerIO performs a light structure scan before the full import:

• tabular files can show detected columns and an initial row count;
• documents can show extracted document structure;
• files that cannot be inspected show an error before the import proceeds;
• the user can remove a file or inspect its preview;
• for supported tabular files, the user can choose which columns to retain.

This structure scan is not the final PII scan.

Step 2 — Import, profile, and build the dossier

The import pipeline creates the working source and builds available intelligence such as:

• source identity and format;
• schema and inferred types;
• row, column, or document-structure information;
• column statistics;
• searchable chunks and embeddings where applicable;
• data-quality signals;
• entities and domain intelligence where available; and
• task and readiness state.

The interface remains responsive while longer work runs in the task system.

The dossier is built from this profiling work. Privacy processing can continue as a separate pipeline stage, so do not present these as one indivisible completion state.

Step 3 — Privacy coverage

Import completion and PII-scan completion are separate states. A source may be imported while its Privacy Scan is still pending, incomplete, or degraded.

Step 4 — Review readiness and correct findings

Open the source detail and privacy views after processing. Review structure, data-quality signals, PII coverage, and findings. Correct false positives or missed values and re-scan when required. A refresh can rebuild the source's current profile; a privacy re-scan refreshes detection and coverage.

Step 5 — Route the task

Local and external model paths

Important distinction: in-app cloud vs. MCP

For an in-app cloud call, RangerIO can mask outbound values and rehydrate supported response values locally.

For an external MCP host, public rehydration is blocked because an MCP tool result is sent back through that host's conversation context.

Current desktop builds

• macOS: Apple Silicon, arm64
• Windows: x64

Startup requirements

Actual working capacity depends on source size, file complexity, selected models, and other applications using memory.

Install on Windows

1. 1Download the current Windows x64 installer for the correct product.
2. 2Run the installer.
3. 3Complete the installer prompts.
4. 4Start RangerIO.
5. 5Complete the in-app setup.

Install on macOS

1. 1Download the current Apple Silicon installer for the correct product.
2. 2Open the disk image or packaged installer supplied by RangerIO.
3. 3Move RangerIO to Applications when prompted.
4. 4Start RangerIO.
5. 5Complete any macOS security prompt and the in-app setup.

First launch

Both products initialize the local data engine and inspect available resources.

RangerIO

• The data engine is available without the bundled llama.cpp runtime.
• Configure an external local provider or cloud provider when model-assisted work is required.
• Provider configuration is separate from file import.

RangerIO Plus

• Includes the bundled local runtime.
• RangerIO-managed local-model setup is available.
• Model readiness is separate from data-engine readiness.

If a model is unavailable, users can still work with the data engine where the selected function does not require that model.

Configure an AI engine

Import, profiling, privacy review, local analytical queries, and standard exports do not all require a generative model. Chat interpretation and AI-assisted reports do require an active engine. Configure engines from Models, not from the Connectors area.

#### Which model option fits your situation?

Lead with where model processing occurs and who operates that AI service.

The Models page has three lanes:

#### Connect a detected local LLM server

1. 1Start the local LLM server and load at least one model in that server.
2. 2Open Models.
3. 3Choose Local LLM Server.
4. 4Review Local LLM servers for detected endpoints.
5. 5Choose Connect beside the running server.
6. 6Confirm that the provider appears in Saved Configurations or as the active model.

If no server is detected, choose Configure beside a known server or enter a custom endpoint, then choose Test & Connect. The local server is a separate application or service; RangerIO does not install or manage it.

#### Connect a cloud/API provider

1. 1Open Models.
2. 2Choose Cloud/API.
3. 3Select the provider type offered by the interface.
4. 4Enter the endpoint when the provider requires a custom endpoint.
5. 5Enter the API key.
6. 6Wait for the provider's live model list, then select a model. If the provider cannot return a list, enter only a model identifier confirmed by that provider.
7. 7Enter a configuration name.
8. 8Choose Test Connection.
9. 9Choose Save & Connect after the test succeeds.
10. 10Confirm the new entry is Active under Saved Configurations.

Provider credentials, charges, retention, and contractual terms belong to that provider.

#### Switch, edit, or remove a saved configuration

1. 1Open Models.
2. 2Find Saved Configurations.
3. 3Choose Activate to switch to a different saved configuration.
4. 4Use the edit control to change its endpoint, model, name, or key, then choose Save Changes.
5. 5Use the delete control only when the configuration is no longer required.

Deleting a provider configuration does not delete imported sources.

#### Use the bundled local runtime in RangerIO Plus

1. 1Open Models.
2. 2Choose Download Model.
3. 3Select a model offered by RangerIO or provide a supported download URL when that control is available.
4. 4Choose Download.
5. 5Follow download progress in the task interface.
6. 6After the model is present, select or activate it from the model list.
7. 7Confirm that the active-model strip reports the intended local model before starting model-assisted work.

Working-data location

RangerIO uses a local data directory under the signed-in user's home directory. The current engine uses RangerIO_local as its local working directory. Imported working data, indexes, metadata, registries, logs, and audit information can be present there.

Updates

The packaged desktop app checks for updates. Automatic download is disabled, so the user is asked before a package is downloaded.

Troubleshooting startup

If RangerIO cannot start a required component:

1. 1Read the displayed readiness or backend message.
2. 2Close memory-heavy applications if the system is low on RAM.
3. 3Retry the action.
4. 4In RangerIO Plus, verify that the selected local model is available.
5. 5In RangerIO, verify that the configured external provider is reachable when a model is required.
6. 6Open Support from the application header if the issue persists.
7. 7Choose Generate Report.
8. 8Review the diagnostic report, choose Copy Report, and send it to RangerIO support only after checking what it contains.

This chapter must be a full in-product workflow, not a format list.

Projects and project groups

A project organizes related sources for an engagement or analysis. A project group organizes multiple projects.

#### Create a project

1. 1Open Data.
2. 2Open Actions.
3. 3Choose New Project.
4. 4Enter a project name.
5. 5Choose Create.

#### Create a project group

1. 1Open Data.
2. 2Open Actions.
3. 3Choose New Project Group.
4. 4Enter the group details.
5. 5Save the group.
6. 6Move related projects into it when needed.

Verified import formats in the current release

Tabular and structured

• CSV
• TSV
• XLS
• XLSX
• XLSM
• JSON
• JSONL
• Parquet

Documents and text

• PDF
• DOCX
• PPTX
• HTML
• TXT

Import one or more files

1. 1Open Data.
2. 2Choose Import.
3. 3Select Files.
4. 4Add files with the file picker or drag files into the import area.
5. 5Wait for the light structure scan.
6. 6Review every file's status.
7. 7Remove files that should not be imported.
8. 8Open a file preview when structure needs confirmation.
9. 9For a supported table, confirm the selected columns.
10. 10Choose a project or create a project from the import flow when available.
11. 11Leave Domain on Auto-detect or choose the correct domain.
12. 12Choose a Reporting Profile when a relevant profile is available.
13. 13Add Key Notes when the future analysis needs context. Example: “Q4 audit data; focus on vendor transactions above $10,000.”
14. 14Choose Start Import.
15. 15Follow progress in the task panel.

The Domain, Reporting Profile, and Key Notes fields guide later intelligence and reporting. They do not change the source data.

Drag files directly onto Data

The Data page accepts file drag-and-drop. Dropped files open the import workflow with those files staged. The user must still review the staged files and start the import.

Import selected files from a directory

This is a selectable directory import, not continuous directory watching.

1. 1Open Data.
2. 2Choose Import.
3. 3Select Directory.
4. 4Choose the directory path.
5. 5Select the file types to discover.
6. 6Choose whether to include subdirectories.
7. 7Set the maximum directory depth when shown.
8. 8Choose Scan Directory.
9. 9Review the discovered files.
10. 10Filter the list by file type when useful.
11. 11Use Select All, Deselect All, or select individual files.
12. 12Choose Continue with N files.
13. 13Follow the resulting import tasks.

Database and API cards

The current import interface may show SQL Database and REST API source cards, but their public import action is not wired as a complete workflow and displays Coming Soon.

Database and REST API sources are delivered through RangerIO Enterprise licensing, together with centralized management and full support for connected sources. Enterprise licensing is expected to begin shipping in early 2027 and is available on request today.

During import

The task system reports current work and can show partial, interrupted, paused, failed, or completed states.

If an import is interrupted after a restart:

1. 1Open the banner on Data or the task panel.
2. 2Choose Resume to continue when resources permit.
3. 3In the Data banner, choose Dismiss to remove the banner without resuming. In the task panel, use Cancel Task when the interrupted task should not continue.
4. 4If Resume is refused because memory is low, close other applications and retry.

What appears after import

A source row and detail pane can show:

• source name and format;
• project placement;
• row or document information;
• indexing or dossier readiness;
• privacy severity;
• privacy coverage;
• data-quality and structure warnings;
• PII count or unscanned state;
• refresh, chat, export, move, and delete actions when available.

The Data toolbar also supports:

• search;
• status/type/severity filters;
• selecting one or more sources;
• Use in Chat for selected ready sources;
• refresh and move actions; and
• deletion.

Source status

Use these concepts carefully:

• Ready: the inputs required by that feature are available.
• Needs attention: a warning, incomplete process, or uncertain structure requires review.
• Failed: the relevant process did not complete.
• Unscanned: the PII scan has not completed.
• Degraded / Unverified: PII findings may be real, but scan coverage is incomplete.

A source can be analytically ready while its privacy coverage still needs attention.

Excel workbooks

A supported workbook can produce related sheet sources and a workbook-level relationship. Users should:

1. 1Open the workbook or relevant sheet in Data.
2. 2Review sheet-level row, schema, quality, and PII information.
3. 3Use the specific sheet when a question or export must be sheet-scoped.
4. 4Use the workbook context only when cross-sheet aggregation is appropriate.
5. 5Verify source attribution in answers and reports.

Refresh a source

1. 1Open Data.
2. 2Select the source.
3. 3Use its refresh action, or select sources and use Refresh selected.
4. 4Follow the task state.
5. 5Re-open the source detail.
6. 6Review structure, intelligence, and PII coverage again.

A refresh re-imports and re-profiles the source.

Delete a source

1. 1Select the source.
2. 2Choose Delete or Delete Selected.
3. 3Confirm the exact source or sources.
4. 4Verify that the correct project remains.

Deletion removes RangerIO's working records for the selected source. It does not delete the original external file.

What a dossier is

The dossier is RangerIO's reusable structured understanding of a source. It is built from import and profiling work and used by source details, questions, reports, and connector tools.

A dossier is not a second copy of the original document for human reading. It is a structured intelligence layer.

What can appear in a dossier

Depending on the source and completed processing:

• source name, format, size, and timestamps;
• row, column, page, paragraph, or document-structure information;
• schema and inferred types;
• per-column statistics such as missingness, uniqueness, common values, and numeric summaries where applicable;
• data-quality and parsing-confidence signals;
• detected entities;
• PII findings and scan coverage;
• searchable chunks and retrieval metadata;
• domain and reporting-profile context;
• readiness, degraded, or blocked state;
• workbook/sheet attribution; and
• provenance used by questions and reports.

Not every field is available for every format.

Open a dossier/source detail

1. 1Open Data.
2. 2Search or filter for the source.
3. 3Select the source row.
4. 4Review the detail pane.
5. 5Open the data or document preview for row-level or text-level context.
6. 6Open the privacy view when PII findings require review.
7. 7Use Use in Chat or the chat action only after the source is ready for the intended workflow.

Readiness contract

A Ready status does not certify the underlying data, the completeness of PII detection, or an AI answer.

When the dossier changes

The dossier can change when:

• a source is imported;
• a source is refreshed;
• structure or intelligence is rebuilt;
• PII findings or overrides change;
• a PII re-scan completes; or
• a source is processed with updated product components.

After any material change, re-open the warnings and readiness indicators before exporting.

Data-quality signals

RangerIO can surface structure confidence, missingness, anomalous values, normalization issues, and other data-quality context depending on the source and operation.

For analytical answers, caveats can be operation-aware. For example, a total over a materially incomplete column can carry a missingness warning.

A missing warning is not proof that the data is clean.

This chapter should contain separate workflows for structured tables and documents.

What the Privacy Scan does

RangerIO scans supported structured and document content for potential sensitive values. Detection can combine machine-validatable patterns, model-detected entities, prior user marks, and stored corrections.

The public review experience uses these labels:

Coverage and severity are different

• Severity describes the risk level of findings that exist.
• Coverage describes whether scanning completed sufficiently.
• Unscanned is not a severity.
• Degraded · Unverified means findings may exist but the scan is incomplete.

Never present zero findings from an unscanned source as “No PII.”

Review a structured source

1. 1Open Data.
2. 2Select the source.
3. 3Open its preview.
4. 4Choose Privacy to open the findings view.
5. 5Review flagged columns and the displayed type/severity.
6. 6Click a PII-marked column header to inspect its current classification.
7. 7
8. 8Mark a value or column as PII, dismiss a false positive, or reclassify the type when the interface offers that action.
9. 9Re-scan after material corrections.
10. 10Recheck the source's coverage state before cloud use or masked export.

Review a document

1. 1Open the document preview from Data.
2. 2Choose Privacy.
3. 3Read the document with scan-backed underlines.
4. 4Use the class chips to focus on:

- Needs your review - Marked by you - Verified

1. 1Use the previous/next occurrence controls to step through every occurrence of a finding.
2. 2Click a marked value to open its correction actions.
3. 3Use:

- Not PII — this value to dismiss a detected value; - Reclassify when the PII type is wrong; - Confirm as [type] for an AI-detected value; - Unmark for a value previously marked by the user.

1. 1Select unmarked text in the document and choose Mark as PII when the scan missed it.
2. 2Use Reading mode when you want the marks reduced to quieter gutter indicators.
3. 3Finish all corrections before running one final re-scan.

Correction scope

A document correction applies to every matching occurrence represented by that correction. The correction view shows the occurrence count. Review the count before applying the action.

Re-scan and export gate

Corrections appear in the preview immediately, but masked document export uses the last complete scan.

After a correction:

1. 1Read the banner explaining that corrections were applied.
2. 2Choose Re-scan now when review is complete.
3. 3Leave the preview open if desired; it remains usable while scanning.
4. 4If re-scan fails, the corrections remain saved but detection has not refreshed.
5. 5Retry the re-scan.
6. 6Wait for Scanned [time] · export ready before using the one-click masked export.

When the scan is stale, the header shows Masked export paused until re-scan.

If the scan found nothing

A completed scan with no findings is not a guarantee. Review the document and use text selection to mark anything missed.

What automated detection cannot promise

• complete recall for every locale and identifier;
• perfect distinction between real and template data;
• correct classification of every organization, date, account reference, or short name;
• a legal conclusion about whether a value is personal information; or
• permission to disclose the resulting file.

Human review remains required.

Purpose

Safe Handoff is the set of controls used when an explicitly configured cloud or connector workflow receives RangerIO context.

In-app cloud-provider flow

1. 1Read privacy state

RangerIO checks the selected source's PII coverage and current registry.

1. 1Build task context

Only the context required for the question or report is assembled.

1. 1Mask or tokenize detected values

Registered values are replaced with stable stand-ins where the workflow supports it.

1. 1Validate outbound content

RangerIO applies its pre-egress checks. If acceptable sanitization cannot be established, the cloud call is blocked.

1. 1Send to the configured provider

The user has explicitly configured this provider and is responsible for its account and terms.

1. 1Receive the response

The in-app workflow can restore supported tokens locally.

1. 1Show lineage and warnings

The response can include protection markings, grounding, caveats, and audit metadata.

Stable stand-ins

Stable stand-ins let repeated values remain linkable during an engagement. This can preserve relationships such as repeated counterparties without sending the original detected value.

The mapping is maintained in a local registry scoped to the engagement/project workflow.

Fail-closed behavior

If RangerIO cannot validate a cloud-bound context, the expected outcome is a blocked cloud request with an explanation—not a silent downgrade.

The user can then:

• re-scan the source;
• correct findings;
• select fewer or different sources;
• use a local provider; or
• use RangerIO Plus's bundled local runtime when available.

External MCP flow

An external MCP host is different:

1. 1The host asks RangerIO to run a tool.
2. 2RangerIO authorizes the tool and builds a response.
3. 3PII-aware scrubbing and egress controls apply.
4. 4The tool result returns to the external host.
5. 5That result becomes part of the host's conversation context.

Because of step 5, RangerIO's public MCP surface blocks rehydration. The external host receives sanitized or tokenized content.

What MCP tools may return

Depending on the tool and scope:

• project and source metadata;
• dossier summaries;
• schema and quality information;
• PII summaries;
• sanitized row previews;
• sanitized analytical query results; and
• tokenized exports when the user explicitly enables that permission.

Review requirement

Masked context can still expose structure, categories, counts, excerpts, or undetected values. The user must decide whether the task and external provider are permitted.

Start a data question

1. 1Open Assistant.
2. 2Choose Select a data source.
3. 3Search for and select one or more ready sources.
4. 4Preview the selected source when needed.
5. 5Enter the question.
6. 6Use @ to reference available fields where useful.
7. 7Send the question.
8. 8Follow progress and use Cancel if the query should stop.
9. 9Review the answer and its evidence signals.

Users can also select sources on Data and choose Use in Chat.

How RangerIO routes a question

RangerIO uses multiple answer paths:

1. 1Direct dossier lookup

Schema, counts, known fields, and precomputed source facts can be returned without a generative model.

1. 1Local analytical query / Smart SQL

Questions that map to a safe analytical query can run against the local analytics engine.

1. 1Grounded model-assisted answer

Interpretive questions use a configured model with source context.

The user does not need to choose the internal route.

Example direct questions

• “How many rows are in this source?”
• “Which columns are present?”
• “Which selected sources are still unscanned?”
• “What is the inferred type of Invoice Date?”

Example analytical questions

• “Show the top 10 vendors by total amount.”
• “Count transactions by status.”
• “List records over $10,000.”
• “Compare missingness in Q3 and Q4.”

Example interpretive questions

• “Summarize the main patterns in this source.”
• “What should I review first?”
• “Explain the largest differences between these two files.”
• “Draft a cautious summary that includes the data-quality caveats.”

Multiple sources

When selecting multiple sources:

• confirm that each source is ready;
• ask a question that identifies the desired comparison or join;
• review source attribution;
• reduce the number of sources if coverage or model limits become an issue; and
• do not assume all cross-source relationships are valid without checking keys and source structure.

Evidence and honesty signals

An answer can display:

• source names;
• confidence;
• processing strategy;
• data coverage;
• a grounding badge for checked numeric claims;
• caveats for missing or anomalous data;
• a clarification request when confidence is too low;
• a warning when a response was shortened; and
• a protection label when values were masked and restored in an in-app cloud workflow.

Teach users to read these signals.

If a cloud request is blocked for privacy

1. 1Read the privacy error.
2. 2Select the blocked source.
3. 3Use Re-scan now when offered.
4. 4Review the scan and corrections.
5. 5Retry only after coverage is current.
6. 6Alternatively use an available local model path.

If an answer is incomplete

• ask a narrower question;
• select fewer sources;
• name the exact field or operation;
• review missingness and caveats;
• refresh or re-scan the source;
• add a related source only when it is genuinely needed; or
• calculate consequential numbers independently.

Limits of external-client answers

RangerIO can validate and sanitize its tool output. It cannot control or validate new prose that an external AI host generates after receiving that tool output.

What the connector is

RangerIO includes a local MCP server that lets an MCP-compatible host request approved RangerIO tools. The server is off by default.

The current guided setup is for Claude Desktop. The protocol may interoperate with other MCP hosts, but this guide must not claim end-to-end support for another host.

Enable the MCP server

1. 1Open Settings.
2. 2Open Connectors.
3. 3Turn on MCP server.
4. 4Wait for the local connector service to start.
5. 5Choose Add Connector.

Configure Claude Desktop

The wizard performs a preflight and configuration flow:

1. 1Ensure Claude Desktop is installed.
2. 2Close Claude Desktop if RangerIO says it is running.
3. 3Review RangerIO's configuration preview.
4. 4Approve the connector installation.
5. 5Let RangerIO write the managed connector configuration.
6. 6Relaunch Claude Desktop.
7. 7Return to RangerIO.
8. 8Confirm the connector status changes to Connected after activity.

The in-product wizard owns the guided setup.

Connector status

• Not configured: the connector has not been installed.
• Pending: configured and waiting for the first connection.
• Connected: RangerIO has observed connector activity.
• Idle: configured without current activity.
• Error: the sidecar or configuration needs attention.

Use Re-test to check the connector and Disconnect to remove the configured connection.

Trust settings

The MCP connector is conservative by default.

Allow tokenized export from stdio agents is off by default. When off, an agent cannot request a tokenized project bundle outside RangerIO's data directory.

When enabled:

• a trusted agent can request that export;
• detected PII is replaced with tokenized/synthetic equivalents;
• the resulting bundle still leaves RangerIO's data directory; and
• the setting should be disabled when it is no longer needed.

Disable the server

Turning the MCP server off stops the local connector service and unregisters the managed desktop connection. Use this when connector access is not required.

Separate standard data exports, AI-assisted reports, and masked document exports.

A. Export standard data

1. 1Open Export.
2. 2Select a source.
3. 3Leave AI Assisted off.
4. 4Review the preview.
5. 5Choose one of the formats offered for that source.
6. 6Configure available PII/masking options.
7. 7Start the export.
8. 8Follow the real progress state.
9. 9Review the completion path and saved/downloaded file.
10. 10Inspect the exported content before sharing it.

Format availability varies by source. The current export system supports structured formats and document/report outputs; the interface is the source of truth for the selected source.

The format picker can narrow the choices further for a particular source.

B. Generate an AI-assisted report

1. 1Open Export.
2. 2Select one or more eligible sources.
3. 3Turn on AI Assisted.
4. 4Review the cloud/local model disclosure shown by the interface.
5. 5Select a report template when available, or choose the report format.
6. 6Add or select relevant discoveries when offered.
7. 7Choose Generate Report.
8. 8Let the report run in the background.
9. 9Review the generated sections and any available content variants.
10. 10Read missing-section, degraded, or source-unavailable warnings.
11. 11Choose Download for the complete draft or Export Report for the selected variants.

Current report formats shown by the product include DOCX, PDF, and Excel where supported.

A generated report must still be reviewed for factual and professional accuracy.

Report readiness

C. Export a masked document from the preview

1. 1Open the document preview.
2. 2Complete the Privacy review.
3. 3Apply corrections.
4. 4Run Re-scan now.
5. 5Wait for export ready.
6. 6Choose Export masked.
7. 7Follow progress.
8. 8Open the exported copy and inspect every sensitive section before disclosure.

D. Export from the PII dashboard

For supported sources:

1. 1Open the PII report/dashboard.
2. 2Choose the source.
3. 3Review findings by type.
4. 4Select the PII types to mask.
5. 5Select the masking style offered by the interface.
6. 6Choose Export Masked.
7. 7Review the resulting copy.

Masking strategies

The standard Export screen currently offers these labels when the selected source supports them:

• None — export as-is: does not mask detected PII. This must never be described as the safe default for a sensitive source.
• **Partial (J\*\*\*n):** keeps limited edge characters and masks the rest.
• **Full (\*\*\*\*\*):** replaces all visible characters in the detected value.
• Redact: removes the detected PII from the visible output.
• Tokenize: replaces registered values with deterministic PII tokens so selected relationships can remain linkable.

Availability and behavior depend on the source and export path.

Audit trail

RangerIO records local audit information for protected operations, including connector and export activity where supported. Audit records are technical evidence of what RangerIO recorded; they are not a regulatory certification.

Organize this chapter first by the job to be done, then by profession. A reader should not have to identify with an industry label to find a useful procedure.

Check a document and create a masked review copy

Ask a spreadsheet question and verify the answer

Compare two files or reporting periods

Prepare a draft report for professional review

Review a folder of engagement files without enabling folder watching

Use RangerIO from Claude Desktop

Individual professional workflow

1. 1Create a project for the engagement.
2. 2Import supported files.
3. 3Review structure warnings.
4. 4Complete the Privacy Scan.
5. 5Correct findings and re-scan.
6. 6Ask local or model-assisted questions.
7. 7Verify important answers.
8. 8Export the dossier, report, or masked copy appropriate to the task.
9. 9Retain or delete the local project according to organizational policy.

Accounting and audit workflow

1. 1Create a project for the client and period.
2. 2Import ledgers, workbooks, and supporting documents in verified formats.
3. 3Review sheet structure and header confidence.
4. 4Inspect missingness, unusual values, and parsing warnings.
5. 5Review PII across the sources.
6. 6Ask precise questions such as totals, counts, status breakdowns, and exceptions.
7. 7Read grounding and data-quality caveats.
8. 8Verify material totals against the underlying rows.
9. 9Export data or a report for professional review.

Legal and HR workflow

1. 1Create a project for the matter or HR review.
2. 2Import supported DOCX, PDF, spreadsheet, or text sources.
3. 3Open document previews.
4. 4Review AI-detected — needs review values first.
5. 5Step through each occurrence.
6. 6Dismiss, reclassify, confirm, or mark values.
7. 7Re-scan after corrections.
8. 8Use the bundled local model in RangerIO Plus or a permitted configured provider for analysis.
9. 9Export a masked copy only after reviewing the export itself.

Healthcare and research workflow

1. 1Create a project for the authorized dataset or study workflow.
2. 2Import supported tabular exports and documents.
3. 3Review the source's structure and scan coverage.
4. 4Inspect names, dates, identifiers, locations, and other potential re-identification fields.
5. 5Manually mark specialized identifiers that the scan missed.
6. 6Re-scan.
7. 7Use local analysis where policy requires it.
8. 8If an external provider is permitted, review sanitized context and organizational rules first.
9. 9Verify all analytical results against the source.

Repeated engagement workflow

1. 1Start a new project for the new period or client.
2. 2Import the current files.
3. 3Confirm that source structure has not changed unexpectedly.
4. 4Review new scan findings rather than assuming prior classifications apply.
5. 5Ask repeatable questions using the same field names and operations.
6. 6Export the current output and record the review.

It is not a regulatory certification

RangerIO and RangerIO Plus provide technical controls and review tools. They do not make an organization HIPAA-compliant, SOC 2-certified, PCI DSS-compliant, or legally authorized to disclose data.

It is not DLP or endpoint security

The products do not monitor all filesystem, email, browser, or network activity and do not enforce organization-wide data-loss policy.

It is not a cloud warehouse

The current desktop products are workstation-oriented. They are not a centralized multi-user warehouse or enterprise identity/access-control plane.

It is not a real-time streaming pipeline

The verified public workflow imports supported files and selected directory contents. It does not provide a documented continuous event-stream workflow.

Database and REST API import are not complete public workflows

Cards may appear in the import interface, but the current release does not document them as complete, usable import paths. Database and REST API sources are part of RangerIO Enterprise licensing, with full support and centralized management, and are expected to begin shipping in early 2027. For more information on our Enterprise version, please contact info@rangerio.com.

It does not certify a source as free of PII

A complete scan can miss values or produce false positives. A user must review the source and output.

It does not guarantee correct AI output

Models can be incorrect, incomplete, or misleading. Grounding and caveat controls reduce risk but do not eliminate it.

It does not control an external AI host

RangerIO controls its own tool response and egress checks. The host's final prose, retention, account settings, and provider behavior remain outside RangerIO.

RangerIO is not RangerIO Plus

RangerIO does not include the bundled llama.cpp/GGUF runtime.