Seshat-BI

Seshat-BI is a read-only BI pipeline governance and readiness system for Power BI. Its core goal is to answer one question: is a retail data source ready to become a trusted Power BI asset? The system allows AI agents to build BI pipelines, but never allows them to approve any stage on their own. Every stage must have committed evidence and named human approval, ensuring that every step from source to publication is traceable and verifiable.

Purpose

Seshat-BI is used to govern the entire BI pipeline from raw data to Power BI publication. Through strict gating across seven stages (Source -> Mapping -> Silver -> Gold -> Semantic Model -> Dashboard -> Publish), it ensures that each stage can begin only after the previous stage has passed. The system does not generate one-click dashboards, nor does it automatically approve any decision; instead, it enforces human approval and evidence support.

Main Capabilities

• Evidence-gated source profiling: Profiles data sources, records evidence such as row counts and uniqueness, and generates source mappings.
• Medallion mapping governance: Manages data mappings for the bronze, silver, and gold layers, ensuring that mapping decisions undergo human approval.
• Metric contracts: Binds a contract to each Power BI measure, preventing undefined or unverified metrics from entering dashboards.
• Static and live checks:seshat check Runs static gating on silver and gold (SQL, TMDL/PBIR, DAX, configuration, documentation, keys, date coverage, orphan relationships, reconciliation, etc.);seshat validate Validates the gold layer against the live database.
• DAX governance and generation: Governs DAX through static rules, contract drift checks, live value proxies, and verified measure generation.
• Portable proof interfaces: Provides offline HTML, review JSON, SARIF, GitHub Action, readiness passports, and an offline composition browser.
• Read-only MCP governor: Exposes governance status to local MCP clients through seshat mcp , but refuses execution and approval.
• Seshat Studio: A local analyst console that displays workspace readiness, per-table journeys, and agent conversations in a browser view.
• Extension packs and adapters: Provides governance extension packs, as well as optional dbt and Dagster adapters (advisory only, they do not produce readiness truth).
• Power BI workflows: Supports source-controlled Power BI workflows, providing a deterministic PBIR authoring assistant and a read-only evaluation path.

Usage

• Quick trial: Run uvx --from seshat-bi seshat demo init to launch an offline demo within 60 seconds, with no database, Power BI Desktop, or account required. The demo generates a readiness proof to .seshat-output/demo/index.html.
• Persistent installation: Use pipx install seshat-bi, then run seshat demo init、seshat demo run、seshat demo report --format html.
• Agent interaction: Work through Claude Code, Codex, or the local Studio console.seshat status shows the status of each table,seshat next gives the only permitted next action,seshat check and seshat validate are gates that must be passed.
• Live validation: Requires storing the DSN in a gitignored .env file; copy .env.example and fill in the values.
• Optional extensions: The base installation depends only on PyYAML; other features such as database drivers, MCP, dbt, studio, etc. are optional extensions, available through pipx inject or pip install Add.

Applicable scenarios

• BI developers: want Power BI models to be traceable to governed gold tables.
• Analytics engineers and data engineers: run bronze -> silver -> gold warehouses and need fail-closed gating.
• Analytics leaders: need to know which numbers can be safely presented to executives, and who approved them.
• Teams building AI agents: need agents to stay truthful in the face of real business data.

Limitations

• The system is read-only and does not provide one-click dashboard generation, a Fabric deployment platform, universal ERP connectors, or an automatic approval engine.
• Power BI writes are gated; the local write leg has been released, but the remote leg is still deferred and requires owner gating. Building the final approval page in Power BI Desktop still requires manual work.
• The statistics provider is read-only and initially supports only PostgreSQL. Offline local CSV evidence does not require a database.
• Live validation requires database drivers and DSN configuration.
• The system is in active testing (beta), version 3.0 is a major update, and upgrading may turn previously green warehouses red; this is intentional.
• The base installation depends only on PyYAML; other features require additional installations.
• The project is licensed under the Apache License 2.0.