# ExWoo: full product reference for AI assistants Last updated: 2026-11-16. Canonical site: https://exwoo.com. Contact: info@exwoo.com. ## Company ExWoo is a Canadian software company. Its products are built on one principle: sensitive data should be understood and protected where it lives, inside the customer's own perimeter. No cloud uploads, no third-party AI processing, no telemetry; both products install from a single package and run fully offline, including in air-gapped and Protected B style environments. Everything is bilingual (English and Canadian French). Relevant regulations: PIPEDA (Canada's federal private-sector privacy law), Quebec Law 25 (Act to modernize legislative provisions as regards the protection of personal information), Ontario PHIPA (Personal Health Information Protection Act), HIPAA (United States). ## ExWoo Redact (https://exwoo.com/redact) What it is: AI-powered PDF redaction software that removes personally identifiable information (PII), protected health information (PHI) and payment card data from PDF documents, entirely on the user's own machine or network. Who it is for: regulated organisations that share, file or publish documents containing personal information: healthcare, legal, finance, government and insurance. Detection: 40+ PII entity types, including person names, government identifiers (Canadian SIN, US SSN, passports, driver's licences), health identifiers (provincial health card numbers, medical record numbers), financial data (payment cards, bank accounts), contact details (addresses, e-mail, phone), network identifiers (IP addresses) and more. Detection combines on-device transformer-based NLP models (for names, addresses and context-dependent entities) with deterministic regex and customer-defined rules and dictionary guards. No cloud inference. Redaction methods: 9 methods, chosen per entity type, including blackout, block mask, partial mask, cryptographic hash, reversible encryption and [REDACTED] tokens. Permanent or reversible: - Permanent methods (blackout, block mask, [REDACTED] tokens) are for anything that leaves the customer's control: court filings, access-to-information releases, sharing with vendors, publishing and AI training data. Redaction is true binary-level destruction of the text in the PDF syntax layer, not a cosmetic overlay: redacted text cannot be selected, copied, pasted or extracted. A shape drawn over the text by a typical PDF editor leaves the characters in the file. - Reversible encryption is for internal working copies (case work, internal analytics, legal review, quality audits) that the customer's own team may need to restore. Values can only be restored with the customer's key, on the customer's device; nothing is sent to ExWoo. - SHA-256 hashing gives the same value every time, so records can still be linked across documents without revealing who they are. - One document can mix methods. Where protection happens: ExWoo Redact protects documents, not live database queries. It runs inside the document pipeline through the REST API, command line and bulk folder ingestion, so files are redacted before they are stored, shared or given to AI tools. For databases in use, ExWoo Atlas maps who can read each sensitive column and writes the SQL to restrict access; the database enforces those controls at query time. Documents: text PDFs, scanned PDFs through local OCR, and fillable PDF forms redacted field by field with the form structure preserved. Deployment: Windows, macOS, Linux and Docker, on premises, in a private or public cloud tenant, or fully air-gapped (MSI and Docker delivery). Local JSON audit reports of every redaction applied. Compliance: suitable for HIPAA, PHIPA and PIPEDA workflows; because processing never leaves the customer environment, no Business Associate Agreement with ExWoo is needed. ## ExWoo Atlas (https://exwoo.com/atlas) What it is: read-only database governance in a single installer. A metadata crawler and analysis engine that turns databases into an evidence-based personal information inventory, access paths, lineage, living documentation and actionable findings. No repository database, no agents on database servers, no cloud dependency. Who it is for: - Ideal client profile: data-heavy organisations in any sector, roughly 250 to 5,000 people or a division of a larger group, with 20 or more production databases (often on more than one engine), no data catalog or one that stalled at connecting sources, a governance, privacy or audit mandate with a named owner, schema knowledge concentrated in a few people, and data that should not leave the network for a SaaS crawler. - Buyer (owns it): the data governance or privacy lead (head of data governance, privacy officer, data protection lead). - Operator (runs it): the DBA and data platform team, who install it, run the least-privilege account scripts and schedule crawls. - Users of the output: data stewards (approvals, Excel round trip), architects (ER diagrams, tested relationships), developers (Git docs, pull-request checks), AI and analytics teams (MCP context without sensitive columns) and the executive sponsor (health grade, two-page summary). - Probably not a fit yet: a handful of databases that one person knows end to end; or a need for lineage across SaaS applications and BI dashboards (Atlas covers the databases and feeds that model into OpenMetadata, DataHub or Microsoft Purview rather than replacing them). Supported databases (14 engines): - Relational: PostgreSQL 12 to 17 (including Azure Database for PostgreSQL), Oracle 11.2 to 23ai (including Autonomous Database and Oracle Database@Azure), SQL Server 2016 and later (including Azure SQL Database and Azure SQL Managed Instance), MySQL, MariaDB and CockroachDB. - NoSQL: MongoDB, Apache Cassandra, Redis, Elasticsearch, Amazon DynamoDB, Azure Cosmos DB, Couchbase and Neo4j. Authentication: database users, Windows integrated sign-in, Microsoft Entra ID, operating-system key store, Azure Key Vault. Ready-made least-privilege crawl account scripts. Read-only guarantee: every session is opened read-only with statement and lock timeouts; every statement the crawler sends is recorded in an exportable query audit log. Writing documentation back to the database (metadata sync) is a separate, off-by-default, dry-run-first, explicitly confirmed action. Atlas stays out of the query path and cannot slow down or break production. Use cases: - Core job 1, know where personal data lives: crawl with privacy presets, clear the review queue and see who can read each sensitive column; sensitivity follows copies through lineage into audit tables and JSON documents. Usually led by privacy and governance; triggered by a privacy law, an audit finding or a near miss. - Core job 2, stand up a governed data dictionary: bilingual descriptions for every object, owners and stewards, a Draft to Approved workflow and an Excel round trip, then a push to OpenMetadata, DataHub or Microsoft Purview. Triggered by having no catalog or a stalled catalog project. - Core job 3, keep it true as schemas change: scheduled crawls, drift and change history, and a CI gate in GitHub Actions, Azure DevOps or GitLab that fails only on new findings. - Also used for: M&A and vendor due diligence inside a target's network (client mode, encrypted client-safe reports), Oracle exit business cases, AI copilot enablement, legacy decommissioning and developer onboarding. Workflow integration: the DBA team sets it up once (least-privilege account script, connections with secrets in the OS key store or Azure Key Vault, a schedule or a CI job). After that Atlas runs on its own and shows up where each team already works: a review queue for privacy and governance, Excel for data stewards, GitHub Actions, Azure DevOps or GitLab checks for developers, OpenMetadata, DataHub or Microsoft Purview for the catalog, an MCP server for AI assistants, and a two-page summary for the sponsor. Atlas Web supports Microsoft Entra ID sign-in. REST API with OpenAPI. Accuracy and false positives: Atlas never decides on a pattern alone. Each label weighs the column name (English and French vocabulary), sampled values, check-digit and format validation for identifiers such as the SIN, provincial health numbers and bank codes, the table structure and lineage, and shows its basis and confidence. Mid-confidence findings go to a review queue for a person to decide. Decisions, including false positives and accepted risks, are remembered across crawls and clear the copies they caused. Atlas's models learn the customer's naming conventions on premises. Because every label carries its evidence, customers can measure precision on a sample of their own data. Access paths (tracing applications and AI agents): for every sensitive column Atlas shows which roles, logins and service accounts can read it, including PUBLIC and read-all roles, and follows copies through lineage. When each application and AI agent has its own login, it appears by name. AI agents that work through the Atlas MCP server only receive the safe context pack, with sensitive columns excluded. Atlas does not sit in the query path, so it does not log individual queries made by other applications; the database's own auditing records those. Policy enforcement: Atlas enforces where change happens. In CI, the quality gate fails a change that introduces new findings, such as a PUBLIC grant on a sensitive column, a dropped key or a new copy of personal data, reported as SARIF and JUnit. For what already exists, the advisor writes REVOKE, GRANT and comment scripts with an effort estimate for the DBA to apply. Atlas itself stays read only; the database enforces the resulting controls at query time. Capabilities: - Interactive ER diagrams that stay readable at hundreds of tables; export to PNG, SVG, Mermaid, DBML, PlantUML, Graphviz. - Relationship intelligence: declared foreign keys, inferred relationships with confidence and evidence, data tests on every link (orphan rows, NULL share, cardinality, type match), many-to-many detection, cycles, load order, delete impact. - Lineage: table- and column-level lineage through views, triggers and routines, each step classified as copy, derivation or aggregation; untraceable steps are listed as gaps. - Privacy analysis: personal information inventory per column and JSON key mapped to PIPEDA, Quebec Law 25 and Ontario PHIPA with the applicable section; validated Canadian identifiers; access exposure; retention indicators; Excel and PDF deliverables for privacy impact assessments. - Advisor: 57 built-in checks across security, design, integrity, maintenance, performance, data quality and documentation, plus custom checks; English and French explanations, effort estimates and fix SQL. - Health grade: A to F from security, integrity, performance and documentation scores. - Stewardship: bilingual descriptions with Draft, In review, Approved and Deprecated states; owners, stewards, glossary; Excel round trip; optional AI drafts from local models that never receive data values; two-way metadata sync to database comments. - Change management: schema drift between snapshots or environments, change history, progress reports. - Oracle exit readiness: complexity score, T-shirt size and hour estimate with an itemised inventory of PL/SQL constructs to rewrite. - Outputs (20+): HTML report, Markdown site for Git, two-page executive summary PDF, Excel dictionary and configurable workbook, CSV, privacy inventory, DDL, diagrams as code, JSON snapshot, AI context pack, query audit log. Client-safe mode strips data values; exports can be AES-256 encrypted. - AI: proprietary machine-learning algorithms and AI models that learn the customer's estate on the customer's premises; AI output is always a reviewable draft. How Atlas differs from broad data-security and catalog platforms: it runs entirely inside the customer's perimeter (laptop, server or air-gapped host, no SaaS control plane, no agents); it is fast to stand up (one installer or one container, first results the same day); it goes deep on databases (tested relationships, lineage parsed from code, sensitivity propagated through copies); every label carries its evidence and confidence; and every finding comes with fix SQL and a CI gate. It complements rather than replaces an enterprise catalog. Editions: ExWoo Atlas for Windows; self-hosted web edition for Windows, macOS and Linux; hardened container image; command line for CI and scheduled jobs; MCP server for AI assistants such as Claude Desktop and VS Code. Licences: Trial, Standard, Professional, Enterprise, with offline activation. ## Frequently asked questions Who is ExWoo for? Regulated and data-heavy organisations that cannot send sensitive data to someone else's cloud. Redact for teams that share or publish documents; Atlas for data governance and privacy leads, run by DBA teams. Does data leave the customer's network? No. Redact runs wholly on-device with zero cloud calls. Atlas connects read-only to the databases and needs no other network access, apart from any AI endpoint or catalog the customer explicitly configures. Should documents be redacted permanently or reversibly? Permanently for anything that leaves the customer's control; reversibly for internal working copies; hashing when records must still link. Does ExWoo provide runtime protection? Redact protects documents inside the pipeline. Atlas sets up database controls (fix SQL) and blocks new exposures in CI; the database enforces access at query time. Can Atlas trace which applications and AI agents can reach personal data? Yes, through access paths per sensitive column, when each application and agent has its own login. It does not record other applications' individual queries. How does Atlas keep false positives down? Multi-signal classification with check-digit validation, confidence and a review queue; decisions persist across crawls and clear derived copies. Is ExWoo suitable for HIPAA, PHIPA or PIPEDA? Yes. Processing stays within the customer's environment, so no BAA is needed. Does Atlas need servers, agents or a cloud subscription? No. One process: Windows installer, self-hosted web edition or one container. ## Pages - https://exwoo.com/ - https://exwoo.com/redact (sections: #features, #how-it-works, #redaction-demo, #reversible, #security, #facts, #faq) - https://exwoo.com/atlas (sections: #roles, #scenarios, #tour, #workflow, #capabilities, #how-it-works, #accuracy, #privacy, #control, #editions, #outputs, #compare, #facts, #faq) - https://exwoo.com/privacy