Everything in the product is reachable programmatically. Ground your own agent, give each of your customers an isolated brain, or enrich a text-to-SQL pipeline — the same primitives are available however you want to consume them.
Call the Brain, Intelligence Traces, and the Wiki directly from your backend — the same primitives the hosted MCP tools use. This is how you embed grounding inside your own product instead of exposing it to an agent.
Scaffold a project, push a schema, run a connector, inspect the Brain — without leaving the terminal. Everything the dashboard does, scriptable.
Pre-built agent recipes for the grounding patterns everyone re-derives: query enrichment, event ingestion, account lookup. Drop them in instead of hand-rolling the same MCP tool-calling loop.
The Brain is served as a hosted MCP server — Claude, Cursor, or your own orchestrator connects and gets grounded tools with no custom integration. A REST API covers everything else.
Working sample repos, not snippets: a text-to-SQL agent grounded against a schema plus docs, a Slack bot that files traces, and a multi-tenant setup with an isolated brain per customer.
github.com/brainbox-labs ↗Building for your own customers, not just your own team? Provision an isolated brain per tenant and pass the tenant at query time — boundaries are enforced by the platform, and every customer gets a real model of their own business. See the multi-tenant solution.
Tag @brainbox in a thread and the message is interpreted against the Brain and written as a trace — the same path as email and meetings. Zero pipeline code.
Grounded answers, not guesses — from a living model of your business.