Pricing that scales with usage, not customers
Every plan includes unlimited customer tenants. You pay for tool calls, never per client.
- 100,000 tool calls / mo, then $1.00 / 1k
- 3 MCP servers
- Connect dbt, Cube, or warehouse
- SQL and semantic queries
- Per-tenant isolation and RBAC
- Custom skills and MCP App visualisations
- Connect your IdP (WorkOS, Clerk, Okta)
- Email support
- 1,000,000 tool calls / mo, then $0.70 / 1k
- 10 MCP servers
- Everything in Starter, plus
- Usage insights and analytics
- Rate limits and feature-gating to resell
- Audit logs
- Priority support
- 5,000,000 tool calls / mo, then $0.40 / 1k
- 25 MCP servers
- Everything in Growth, plus
- SSO and SCIM
- Volume discounts
- 99.9% uptime SLA
- Dedicated Slack support
- Volume-based pricing
- Unlimited MCP servers
- Everything in Scale, plus
- Self-host: VPC, on-prem, air-gapped
- Dedicated tenancy
- Custom SLA and security review
- Dedicated support
7-day free trial on all plans. No credit card required.
Frequently asked questions
Setup and data
Do I need a semantic layer to start?
No. Connect a semantic layer (Cube, dbt Semantic Layer, AtScale, LookML) and we import it. Have only dbt models or raw warehouse tables? We read your dbt manifest, or you describe tables and define metrics in Bonnard.
How do I choose what data my customers access?
You expose curated views, fields, and metrics, not raw tables. Anything you do not expose is invisible to the agent. Field- and row-level governance then controls what each customer and user sees within that.
Which warehouses and tools do you work with?
Snowflake, BigQuery, Databricks, Postgres, Redshift, and ClickHouse, plus dbt and Cube. Your data stays where it lives, with no copies.
The MCP
Can I customise my MCP tools?
Yes. Choose which tools each MCP exposes, and keep it read-only or enable a controlled SQL tool. Turn your governed metrics and saved queries into named, parameterized tools of your own. And scope different tool sets to different audiences. The core discover, query, and visualize tools are built for agents and governed by default.
How do agents query my data?
Over a semantic layer, agents request governed metrics, dimensions, and time granularities as a structured JSON query, so results stay consistent and validated. Over raw tables, they can use a governed SQL query. Bonnard supports both and uses the right one for your setup.
Why governed metrics instead of letting the agent write SQL?
Text-to-SQL against raw tables is wrong roughly a third of the time on real schemas. Answers come from the metrics you define, so the number is the number. A controlled SQL path is available for the edge cases.
Which agents and clients can connect?
Any MCP client. Claude, Cursor, ChatGPT, Gemini, and others connect to the same endpoint.
Identity and tenancy
How do you implement tenancy and isolation?
Each customer authenticates through your auth, gets an identity scoped to one tenant, and that scope is enforced at the data layer on every request. One identity, one tenant, never crossed.
What auth methods do you support?
OAuth 2.1, API keys, and bearer tokens. Connect your existing IdP: WorkOS, Clerk, Auth0, Okta, Microsoft Entra, or Google. Customers authenticate as themselves, with no separate accounts.
Can different users within a customer see different data?
Yes. Role-based access controls what each user inside a customer can reach, down to fields and rows. Finance sees finance, support sees support.
Deployment and security
Where can I deploy it?
Managed by us, in your VPC, on-prem, or air-gapped, as a Docker image, on Kubernetes, or via Helm. Deploy in any region for data residency.
Does Bonnard see my data?
No. We never store or train on your data. In managed cloud, query results pass through the gateway in memory and are not retained. Self-host keeps everything inside your own boundary.
How do I see what customers are doing?
Every request is attributed to a customer and identity, with an exportable audit trail and per-tenant usage.