Metadata-only by design. Hosted on AWS.

MyDataWork is a workspace for data practitioners to organize analytical work across their existing tools, connect it to use cases and outcomes, and understand relationships between the assets involved. It is not a data runtime, transformation platform or replacement for enterprise governance. It is an organizational and productivity tool — not a data processing or transformation platform. It does not query, transform, or move data within your systems.

Only structural metadata — never the data values inside your tables or the underlying contents of your files. It falls into three layers:

• Filesystem metadata (what the OS already exposes): file names, paths, types and tool classifications (Excel, SQL, Python, Jupyter, Alteryx, Tableau, Power BI, CSV, ThoughtSpot, Dataiku exports, Looker LookML), sizes, and creation/modified dates.
• Shallow structural metadata (from a file's structure, never its values): SQL FROM/JOIN table names; Excel sheet names; CSV column names; embedded SQL in Jupyter; for Alteryx (.yxmd/.yxwz/.yxmc) the input/output files & tables, tool and ML-model inventory, and join keys; ThoughtSpot .tml object kind and referenced tables; Dataiku project name, recipe/dataset counts, and dataset table references; Looker .lkml dimensions, measures, and sql_table_name. The asset detail panel surfaces this extracted structure read-only — it never displays cell values.
• User-entered content: tags, notes, use-case descriptions, objectives, progress measurements, stakeholder info, optional HTTPS notes URLs, and analyst-defined relationships.

For locally-scanned files, extraction happens on your own machine and only metadata leaves it. For cloud sources, file source content / warehouse INFORMATION_SCHEMA / read-only BI metadata is fetched briefly, parsed in-process, and discarded — only the resulting structural metadata is persisted. When AI features run, a subset of that metadata (never file contents) is sent to OpenAI. The desktop Connector is code-signed by MyDataWork, LLC through Azure Trusted Signing so Windows verifies the publisher at install.

All workspace data is stored in a managed PostgreSQL database (AWS RDS) in the United States (us-east-1 / Northern Virginia). It is encrypted at rest with AWS KMS, has automated daily backups (7-day retention) and point-in-time recovery, and is not publicly accessible — connections are accepted only from within the application VPC.

MyDataWork minimizes the personal data it holds and isolates it from your business data. Three categories, all user-entered:

• Account & user identity: name, email, and a PBKDF2-SHA256 hashed password. Authentication uses session tokens valid for 7 days. Card data is never stored — Stripe processes payments.
• Team member identity: when an admin adds a member, they enter name, email, and a starter password the member changes on first login.
• Stakeholder & contact details: names, optional emails, and any HTTPS notes URLs recorded against a use case.

The core boundary: PII inside your business data is never ingested — connectors read structural metadata only, never file contents, cell values, rows, or query results.

Subprocessors that touch personal data: AWS (hosting, us-east-1; KMS at rest, TLS 1.2+ in transit); Stripe (payments — card data never reaches us); AWS SES (transactional email); OpenAI (AI features — stakeholder names/emails stripped before any call; what reaches the model is role labels, presence/counts, and free-text you author); and Jira (optional — stakeholder and asset names when you push a use case).

Hashing where identity isn't needed: the free-assessment eligibility log and the account-deletion audit log store a SHA-256 hash of the email, never the plaintext address. Full JSON export any time (GDPR Article 20), immediate self-service deletion (erasure), no sale of data, no advertising. DPAs available for organizational customers.

MyDataWork does not connect to, query, or read data from your databases, warehouses, or BI tools — it never executes queries against business data or accesses data values. It can optionally connect to twelve cloud sources using read-only credentials you provide, reading structure only. (Those twelve, plus the local file formats below, are the sixteen sources MyDataWork normalizes into one comparable shape.) What each connector reads:

The local Windows Connector recognizes files by extension — Excel (.xlsx/.xlsm), CSV, SQL, Python (.py/.ipynb), Tableau (.twb/.twbx), Power BI (.pbix/.pbit), Alteryx (.yxmd/.yxwz/.yxmc), ThoughtSpot (.tml), Dataiku exports (.zip), Looker LookML (.lkml), and columnar data files (.parquet/.pq, .avro, .orc) — capturing filesystem metadata plus shallow structural references. For Parquet and Avro it reads the schema from the file footer (column names and types); no data pages are opened. ORC files are catalogued by identity only, because ORC stores its schema in a protobuf footer with no standard reader. Across both local and cloud paths, data values, formula/cell contents, rows, and query results are never read or transmitted.

Minimum permissions: GitHub repo read; Databricks workspace read plus Unity Catalog read; dbt Cloud read on projects/runs/artifacts; Snowflake SELECT on INFORMATION_SCHEMA only; BigQuery Metadata Viewer; Redshift read-only on system catalogs; Power BI read-only admin (service principal, Scanner API); Tableau Metadata API (PAT); Sigma/Looker/ThoughtSpot read-only metadata API; Dataiku read-only DSS key. Cloud connections are optional and admin-configured.

MyDataWork uses OpenAI's API for use-case proposals, recommendations, automation-candidate identification, marketplace-data suggestions, tool-migration analysis, the workspace Assistant, and the Workspace Agent. What is sent: use-case titles, descriptions, objectives, and progress notes; asset names, file types, and detected data-source references; and workspace plan/config. Stakeholder names and emails are stripped before any call — only role labels, presence/counts, and user-authored free-text reach the model. File contents are never sent.

The Workspace Agent runs only when a user clicks “Analyze” — not autonomous, not scheduled. Detection is deterministic business-rule logic inside MyDataWork; OpenAI is used only to phrase the resulting finding in natural language. The Assistant knows your workspace structure but never accesses file contents. Solution Packs and Initiatives use no AI at all. AI is enabled by default (admin can toggle in Setup → AI Assist), subject to a daily quota shown in Setup with the cost of each action disclosed at click; the first Propose run and up to two Workspace Agent attempts are free, and a failed call auto-refunds the credit. OpenAI's data-processing terms apply.

The Asset Estate Assessment is an on-demand AI review; the first run is free on every plan (later runs use credits). It uses OpenAI like other features — file contents are never sent. What's distinct:

• What's sent: asset metadata (names, paths, tool types, topic tags, lineage edges) and use-case text (names/emails stripped), plus lineage-derived structural observations (cross-use-case asset concentration, external dependencies, orphan assets) and a snapshot of recent Workspace Agent findings.
• Workspace-grounding filter: every LLM finding is checked against your actual assets, use cases, and stakeholders by a deterministic filter before it surfaces; template phrasings that name no real entity are dropped. Implemented in the app, not the LLM call, so it can't be bypassed by prompt injection.
• Hashed-email eligibility log: the free-first-run affordance is keyed on a SHA-256 hash of the lowercased email — only the hash and a timestamp persist, never the address. This is the one place a per-email record persists across workspaces; a reviewer sees hex strings, not emails.
• Result storage: the rendered five-section report is stored scoped to your org_id; LLM intermediate outputs are not retained. Disabling AI hides the Assessment entirely.

Architecture & Value (Team and Enterprise plans) is a deterministic rollup — it uses no AI and sends nothing to any AI service. It reads only metadata you already hold in MyDataWork (tools, use cases, recorded value, utilization) and computes the view in-app.

• Aggregates only — the view and its leadership export present rolled-up figures (value by function, by tool, and change-impact), never raw data values or file contents.
• Shared by default — it opens counting only work a member has shared to the team; private work stays private. An admin can opt into a full-workspace view that includes every member’s work, and every such access is written to a durable audit log (who, when, at what scope). A non-admin request for the full-workspace view is refused and narrowed back to shared.
• No cost data — it shows value and utilization only; it never ingests or infers spend.
• Stakeholder handling — where the leadership view or export names stakeholders, those are the people an admin manages in the workspace; stakeholder names are never sent to AI or exposed over agent access (MCP), consistent with the rest of the product.
• Member transparency — members are told at the point of (not) sharing that an item they keep unshared is still counted in the admin’s workspace-wide totals (value and usage only) under this logged view, while the item itself and its contents are never shown.

Agent access is an optional feature that lets a user expose their own workspace context to an external AI agent over the Model Context Protocol — designed so MyDataWork never becomes a data-access or credential broker:

• Opt-in per use case — nothing is shared unless a user grants a specific use case to a specific key.
• Scoped per key — an agent sees only the use cases granted to the key it presents.
• Metadata only — asset names, tool types, structural schema, lineage, and use-case outcomes. Data values, file contents, and stakeholder names are never exposed.
• Map, not keys or data — MyDataWork provides only the map of where data lives (system, schema, table); the agent reads any underlying data through its own connections.
• External data — suggestion and pointer, not a pipeline. MyDataWork recommends public datasets from your metadata alone and generates a connection recipe you run in your own environment. It never joins external data to yours, never executes the recipe, never stores the result, and never sends any of your data — asset names, schema or values — to a third-party API. Source previews fetch public data only and are disabled entirely in fenced demo workspaces. Accepted sources are catalogued as metadata-only assets and are excluded from plan asset limits, on interactive add and on workspace re-import alike. Administrators choose which public sources may be scanned in Setup → Public data sources.
  
  Audited, visible & revocable — every key creation, grant change, and read is logged, and the record is shown back to the user in an Agent activity panel (reads over 7/30/90 days, per-day chart, which key, which tool called, recent-activity feed). Deleted keys retain their history. The activity record is scoped to the key’s owner: in a multi-member workspace one member cannot read another member’s key labels, granted use cases, or tool arguments. Keys revoke anytime.

Access keys are bearer credentials, stored hashed at rest and presented over TLS. Two modes: Live (on-demand reads; paid plans; daily read cap; needs a bearer Authorization header — e.g. Claude Code, custom agents) and Export (a point-in-time metadata file; all plans; works with any agent, including OAuth clients like Claude Desktop). Explorer offers Export only. These caps are separate from AI credits.

Agent Studio helps you scope agentic use cases before they're built. It does not build, run, or route agents and introduces no runtime data path:

• Metadata only, never invoked — read-only capability discovery reads what a server advertises (tool/resource/prompt names, descriptions, input schemas). The client hard-refuses the invoking methods (tools/call, resources/read, prompts/get, resources/subscribe) at the protocol layer, so cataloging cannot cross into execution.
• No credentials or data stored — per tool: name, description, input shape; per source: server name/version, protocol version, auth mode, declared capabilities, counts, timestamp. No secrets, contents, or values.
• AI drafting is metadata-bounded — only tool/source names and labels you type reach the model; covered by a CI PII-boundary test.
• Upstream of your control point — no runtime footprint; not a control point and does not attest a server is safe. Every exported definition ends with a “Before you build” hand-off to route runtime access through your governed control point. Cataloged sources are workspace-scoped and personal by default.

The Jira integration stores your Jira instance URL and API token encrypted at rest in your workspace, using the same encryption applied to all other credentials. They're used only to push use-case summaries to your Jira instance and are never transmitted elsewhere. The content pushed contains only what you entered — use-case titles, descriptions, objectives, progress notes, stakeholder names, and asset names. No file contents, no underlying data, no other system credentials. The integration is optional, disabled by default, admin-configured in Setup → Integrations, and rate-limited to 20 calls/hour.

The Connector is a lightweight Windows desktop app. It typically does not require administrator privileges, does not open inbound ports, does not modify system files, and does not require firewall changes. It communicates only with the MyDataWork web app over HTTPS (port 443) and scans only the folders the user assigns.

No. It asks the Windows Connector (on the user's own machine) to open the file in its default application — the same local action already on the Assets tab. Nothing is uploaded, and MyDataWork gains no new data access. Cloud assets have no local file and deep-link in-app. The Workspace view is read-only over the existing catalog.

Outbound HTTPS to app.mydatawork.com on port 443. No VPN, no special network configuration, no inbound connections — the application is fully cloud-hosted. If the optional Jira integration is configured, outbound HTTPS to your Jira instance on 443 is also made. Rate limits protect platform performance: 5 lineage rebuilds/day, 20 Jira calls/hour, 10 exports/day.

Yes. Team plans designate a single workspace admin (Owner) who manages access — the admin creates member accounts directly (name, email, starter password). There is no email-invite flow into an existing workspace and users cannot self-join one; individuals can self-register at app.mydatawork.com to create their own Explorer workspace only. Seat limits: Team Starter 2–5, Team Growth 6–10. Admins can remove members, transfer ownership, or delete the workspace; a removed member's contributions remain under the admin's control.

Team workspaces are private-by-default. Each member's assets, use cases, initiatives, lineage, stakeholders, AI recommendations, and insights are private to them — including the admin's own work. The default view on every surface is filtered to what the viewing member can see: their own plus assets explicitly published via Share to Team.

Only assets are shareable — use cases, initiatives, lineage, stakeholders, recommendations, and insights stay personal. Windows-Connector imports are attributed to the importing member and stay private until shared. Copies made via “Copy to mine” are independent. Exports (PDF/PPTX) respect the same private-by-default scope. Admins additionally have an opt-in, logged Workspace view of the full catalog for onboarding/audit. No cross-tenant access — each org is isolated at the database level via an org_id foreign key on all tables. The Asset Health Dashboard reflects the same scope.

Yes — from the Account section at any time. Deletion is immediate and permanent, with no 30-day retention (that applies only to subscription cancellation). Team admins must transfer ownership first. All deletion events are logged with hashed identifiers, not full email addresses.

Yes. Full workspace export (JSON) any time from Setup → Data portability — all assets, lineage edges, stakeholders, use cases (objectives, baseline/current/target values, priorities, effort, target dates, notes URLs, communication logs), saved AI recommendations, action plans, and Workspace Agent finding history. Re-importable into any workspace (a safe merge — updates existing, adds new, deletes nothing). Portfolio exports as PowerPoint or PDF. On cancellation, data is retained 30 days for export/resubscription.

The JSON export is suitable for self-managed backups — export anytime, store it in your backup system, and restore by importing (a merge, so re-import is safe). MyDataWork also maintains operational backups: automated daily snapshots with point-in-time recovery and a restore procedure that is documented and periodically tested. Your JSON exports add an independent layer that doesn't depend on our availability.

No. Enterprise changes commercial terms only — negotiated seats, higher per-org limits (assets, AI credits, agent access), and invoice billing. The data handling, member isolation, connector behavior, and AI-processing posture on this page are identical on every plan.

SOC 2 is on our roadmap; our infrastructure runs on AWS, which maintains SOC 2, ISO 27001, PCI DSS, and more. We follow defense-in-depth practices:

• Encryption at rest (AWS KMS) for the database and per-customer integration credentials; TLS 1.2+ in transit with enforced HTTPS.
• Application secrets in AWS Secrets Manager, resolved at container start via a scoped IAM role — never in plaintext config.
• Network isolation: the database is not publicly accessible and accepts connections only from the application. The application itself accepts inbound traffic only from the load balancer, enforced by a security group — it cannot be reached directly.
• Backups & recovery: automated daily backups with 7-day retention, point-in-time recovery, and a documented, periodically-tested restore procedure.
• Continuous monitoring with alerting on application errors, database health, and infrastructure saturation.

Passwords are never stored in plaintext — they're hashed with industry-standard PBKDF2-SHA256. Authentication uses session tokens valid for 7 days. Users reset passwords via email verification. We don't store payment card information; Stripe processes payments.

We support the key rights: portability (full JSON export satisfies GDPR Article 20), erasure (immediate, permanent self-service deletion), and data minimization (we store only what you enter and metadata you choose to index). We don't sell data or use it for advertising, and DPAs are available for organizational customers.

There's no lock-in. Export your complete workspace as JSON any time, or export portfolios and use cases as PowerPoint/PDF. Cancel from Account settings — data is retained 30 days. For immediate, complete removal, use self-service account deletion. The Connector uninstalls like any standard Windows app.

Yes. MyDataWork for Teams is listed in AWS Marketplace as an annual contract for up to 20 users. For many organisations this is the shortest procurement path: the purchase runs through your existing AWS agreement and appears on your existing AWS invoice, so there is no new vendor to onboard, no new payment method, and no separate security questionnaire cycle for a new billing relationship.

AWS acts as seller of record and collects payment. Larger deployments, longer terms, or custom terms can be issued as a private offer in AWS Marketplace.

No. It is the same application, in the same AWS region (us-east-1), with the same metadata-only boundary, the same encryption, the same isolation model, and the same subprocessors described on this page. Nothing about how your data is read, stored, or protected differs by purchase channel.

What differs is billing and entitlement: seats, renewal and cancellation are managed in your AWS Marketplace console rather than in the app’s billing page, and the app checks your entitlement with AWS rather than with Stripe.

AWS sees the commercial facts of the subscription, not your content. That means the purchase itself, the entitlement attached to it (seats and contract term), whether the subscription is active, and the resulting charges on your bill. AWS also provides us with standard seller reports identifying the purchasing account.

AWS does not receive your catalogue, assets, use cases, lineage, stakeholders, or any metadata you manage in MyDataWork. Note that AWS already hosts the service, so AWS is a subprocessor either way — buying through AWS Marketplace does not widen what AWS can see.

Email contact@mydatawork.com. We aim to respond to security-related inquiries within 2 business days.