Semaphor vs Metabase

Two different starting points

Metabase is open-source internal BI at its core, and it is good at that job. Embedding came later, and the pricing and permission model still reflect the internal-BI origin: users sign in, users are billed, and security is organized around groups.

Semaphor was built for customer-facing embedding from the start. Multi-tenancy, white-labeling, fully interactive embedding, and query-time security are core architecture, not tier upgrades. The same governed core also serves internal BI, so your team and your customers see the same numbers.

Pricing

Metabase prices per user on every paid tier, and the detail that surprises teams embedding at scale: end users who sign in through SSO-backed embedding or the React SDK count as billable users. Your bill grows with every customer you onboard, and only view-only guest embeds escape it. Meanwhile the features embedding actually needs (the React SDK, SSO embedding, row security, white-labeling) all sit in the upper tiers.

Semaphor uses flat platform pricing with no per-user fees. Embedding, multi-tenancy, and white-labeling are included in all paid tiers, so your cost does not increase as your team grows or as you roll analytics out to more customers, and the free tier lets you validate before spending anything.

Embedding

Metabase offers three embedding modes: guest embedding (signed JWT iframes, view-only, all tiers), full-app embedding (the whole Metabase app in an iframe with SSO, Pro/Enterprise), and a React SDK for component-level embedding (Pro/Enterprise). The SDK is GA and genuinely component-level, further along than most BI vendors, with documented constraints: React 18/19 only, no server-side rendering, and one dashboard per page.

Semaphor embeds through JWT-signed URLs on all paid tiers, and unlike guest embeds, they are fully interactive: filters, drill-downs, and self-service exploration, white-labeled, in any framework. Past the dashboard, the Data App SDK builds complete governed data apps in your own React code, custom visuals plug in as React components, agents build over MCP, and briefings deliver what changed on a schedule.

Multi-tenancy and security

Metabase implements tenancy with groups and user attributes on Pro and Enterprise. Row and column security filters a table by a user attribute, or swaps the table for an admin-owned SQL question. The official documentation notes the limits that matter for customer-facing use: security applies only to query-builder queries, so users with native SQL access can bypass it; a user cannot be in two row-security policies on the same table; and public links bypass security entirely.

Semaphor enforces security policies when the query is built, before SQL reaches your database, regardless of how the query originates. There is no SQL-editor bypass. Tenant isolation comes in three levels on all paid tiers: row-level (filter by tenant), schema-level (schema per tenant), and connection-level (separate databases per tenant).

Beyond isolation, Semaphor gives tenancy a management model. 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. In Metabase, the equivalent is assembled from groups, user attributes, SSO, and collections, and appearance settings remain instance-wide: one look for every customer.

Semantic layer

Metabase has Models and Metrics: curated datasets with descriptions and field types that act as starting points for exploration. It is a useful curation layer, but not a governed entity layer; there is no relationship graph that resolves joins for end users.

Semaphor has Semantic Domains: datasets, fields, relationships, calculated metrics, and calculated dimensions defined once and shared everywhere. End users drag fields from any dataset and Semaphor resolves the joins automatically, including multi-hop joins. This matters in embedded contexts because your customers do not write SQL; they pick fields, and the system does the rest. Domain setup is AI-assisted: labels, aggregation defaults, and relationships are proposed for you.

Version control

Metabase stores the last 15 revisions of each dashboard. On Pro and Enterprise, serialization exports collections and dashboards as YAML for git-based staging-to-production workflows, with documented gaps: users, groups, permissions, and alerts are not exported, and source and target instances must run the same major version.

Semaphor integrates with GitHub directly. Connect a repository once, and a Commit button appears in the dashboard toolbar. Every change is tracked with author, timestamp, and message; restore any commit from the history. There is no version limit, you review diffs in GitHub, and dashboards can be promoted across dev, staging, and production projects via the API.

AI and agents

Metabot, Metabase’s assistant, creates charts from natural language, writes and fixes SQL, and answers questions in Slack. The managed AI service is Cloud-only; self-hosted instances bring their own LLM key.

Semaphor is agent-native rather than assistant-equipped. An MCP server exposes the governed platform to any agent: Claude or Codex can discover your semantic model, build dashboards and data apps, and verify every number against a live governed query. The same security policies apply to agent-built queries as to human-built ones, and briefings deliver what changed to your inbox without anyone asking.