Semaphor vs Sigma

Where the architectures differ

Sigma and Semaphor share a starting point: both connect directly to your cloud warehouse and run live, governed queries there. Sigma adds a deliberate constraint on top: it never stores anything, so every interaction, every filter change, and every viewer click is a fresh warehouse query. Paired with a spreadsheet interface that encourages constant exploration, that is a lot of queries.

Semaphor differs in two ways that matter as usage grows. Reach: governed connections cover operational databases (Postgres, MySQL, SQL Server, ClickHouse, DuckDB) as well as the cloud warehouses, so analytics is not gated on everything landing in Snowflake first. And result caching: repeated views of the same governed query serve from cache instead of re-running warehouse compute, which matters when hundreds of embedded viewers open the same dashboard every morning.

Pricing

Sigma licenses per user by account type, embed users are licensed under a separately negotiated arrangement, and buyers report usage-based credit charges for input-table updates, API calls, and exports on top. Your warehouse compute bill is additive to all of it, and in an embedded product that compute scales with viewer activity you do not control.

Semaphor uses flat platform pricing with no per-user fees, and result caching keeps repeat viewer traffic from re-running warehouse queries. Embedding, multi-tenancy, and white-labeling are included in paid tiers, and the free tier includes embedding, so you can validate the integration before paying anything.

Embedding

Sigma’s embed stack covers the essentials: server-signed JWTs, a claims reference, an embed sandbox for testing, a React SDK layered over the JS Embed API, and an embeddable assistant. Custom themes and fonts cover white-labeling.

Semaphor embeds the same way: a JWT-signed URL, fully white-labeled, in any framework. The difference shows up past the dashboard: custom visuals are your own React components published through the plugin system, the Data App SDK composes governed queries and visuals into complete data applications, agents build dashboards and data apps over MCP, and briefings deliver what changed, why, and where to look next, on a schedule.

Multi-tenancy and security

Sigma handles tenant isolation with user attributes and teams: JWT claims set attributes on auto-created embed users, and row-level security keys off those attributes. Attributes can also switch the warehouse role per team, reusing RLS you defined in the warehouse. True separated tenant organizations (Sigma Tenants: separate users, connections, and auth under one parent) exist but are a premium public beta you have to request access to.

Semaphor ships three levels of tenant isolation as generally available product: row-level filters, schema-per-tenant, and connection-per-tenant. Policies are enforced when the query is built, apply to every query path including agent-generated ones, and are available on all paid tiers.

It also ships the management model around that isolation. Organizations, tenants, and tenant users are first-class objects: each tenant carries its own users, roles, groups, sharing boundaries, and defaults, down to per-tenant fiscal calendars and number formats, and organization-level appearance controls white-label the whole surface. Tenant users build and share dashboards strictly inside their tenant, enforced server-side. That is the structure Sigma Tenants is working toward in beta; in Semaphor it is shipped product on paid tiers.

Version control

Sigma keeps unlimited workbook version history with tags for dev, staging, and production promotion. But integrating that with real source control is explicitly a do-it-yourself exercise against the REST API; there is no git integration and no as-code format for workbooks.

Semaphor connects to GitHub with a one-click App install. Commit from the dashboard toolbar, review diffs in GitHub, restore any version, and promote dashboards across environments via the API.

AI and agents

Sigma Assistant answers natural-language questions grounded in data models, and Sigma ships an MCP server, with a notable operational detail: you must bring and configure your own LLM provider (OpenAI, Azure, Gemini, Bedrock, or warehouse-hosted models).

Semaphor is agent-native end to end. The MCP server exposes the governed platform to Claude, Codex, or any agent: discover the semantic model, run governed queries, build dashboards and data apps, and verify every number against a live query. Security policies apply to agent traffic exactly as they do to humans, and briefings deliver what changed on a schedule without anyone asking.