Semanticus: the semantic-model workbench for Fabric and Power BI

Semanticus brings semantic-model authoring, analysis and proof into one VS Code workbench. Use the Model tree, Properties and Studio yourself, then let your own AI Assistant work through MCP against the same live session. Every change is attributed, broadcast and placed on one undo timeline.

Semanticus 1.0.1 corrects the Marketplace listing and adds matching-host packages for five operating-system and architecture targets. Every package bundles its engine and is published only after that engine is extracted and executed on the matching CI runner, plus the documented human acceptance checks.

What is included

• Model authoring. Edit supported BIM, PBIP and TMDL models, model properties, DAX, M, relationships, calculation groups, perspectives, roles and modern calendars.
• Best Practice Analyzer. Run compatible rules, apply individual fixes, load custom rulesets and record honest waivers without hiding the finding.
• AI Readiness. Grade model metadata from A to F, explain each finding, apply deterministic safe fixes and prepare descriptions, synonyms, Q&A and Prep for AI metadata.
• Lineage and impact. Trace model dependencies and supported local report references with explicit safe, unsafe or unknown verdicts.
• DAX Lab. Query, benchmark, profile and prove a candidate rewrite against known-good values before applying it.
• Tests and evidence. Store model checks, run deterministic proofs and export tamper-evident HTML and JSON evidence.
• Workflows and Change Plans. Enforce reviewable steps and checks, then apply an approved batch as one undoable transaction.
• Compare and deploy preparation. Compare two models, copy supported objects and preview deployment changes before any confirmed live write.

Supported platforms

The production package is platform-specific and bundles its engine. Users of an accepted package do not need a separate .NET runtime. Remote XMLA and Fabric features still require a compatible tenant, capacity, permissions and credentials. Power BI Desktop discovery, local XMLA and local M preview remain Windows 11 x64 only.

AI-native by design

Run Semanticus: Connect AI Assistant to configure the current workspace for a supported MCP client. The assistant uses the same engine session shown in the UI, so its changes appear immediately and can be undone by the human.

The engine holds no model-provider API keys, performs no inference and sends no product telemetry. AI work happens through the user's own assistant and account.

Free and Pro

The Free tier provides the workbench and individual edits. Pro adds reviewed bulk changes, enforced workflows, advanced evidence and continuous health features. Current plans, terms and account options are published on the Semanticus website; this README does not make a separate price promise.

Licenses verify offline. The extension opens the Semanticus account website for upgrade or management and does not implement checkout, billing, cancellation or license minting.

Quick start

1. Install the platform-specific extension package accepted for your operating system.
2. Open the Semanticus activity-bar view.
3. Run Semanticus: Open Model and select a supported local model, Power BI Desktop instance or XMLA endpoint.
4. Open Studio for readiness, diagrams, lineage, Tests, workflows and deployment preparation.
5. Optionally run Semanticus: Connect AI Assistant for dual-drive work through MCP.

Known limitations

• Power BI Desktop discovery and local M preview are Windows-only.
• Modern calendar metadata requires compatibility level 1701 or later. Older models keep the classic Date-table path until the user explicitly upgrades and authors a calendar.
• Role security is static-first; there is no view-as-role impersonation proof.
• Cloud writes are deliberate: report discovery, Fabric git commit and update, and Data Agent publishing always preview as a dry run first and write only on explicit confirmation.
• When a model or tenant does not support a capability, Semanticus refuses with a clear reason instead of reporting success.

Privacy

Model work runs locally. Network access occurs only for data or package sources the user explicitly opens, the account website the user chooses to open, and supported optional downloads. Semanticus sends no product telemetry.