DigitalKin built a multi-tenant agentic mesh on SurrealDB with live updates in place of polling
DigitalKin builds an agentic mesh: a network of AI agents that work for enterprise customers, each with its own data and its own dashboards. The mesh's document, relational and real-time state runs on SurrealDB, with tenants isolated by namespace.
Challenge
DigitalKin's agents work inside customer environments, so every customer's data has to stay within its own boundary. Spinning up a cluster per tenant would have kept the boundaries but multiplied the cost and the operations with every customer signed.
The mesh is live by nature. Agents hand work to each other and enterprise users watch the results on dashboards, and polling for changes made those dashboards laggy and the webhook chains between agents brittle.
Agent workflows also change quickly, so the data model had to be schema-optional. And the team knew it would want graph relationships between agents, tasks and knowledge, and needed a path to them that was not a rewrite.
Solution
SurrealDB's namespaces gave DigitalKin tenant isolation on one shared cluster, with the same behaviour on SurrealDB Cloud and on self-hosted nodes. Each tenant's data sits in its own namespace with its own permissions.
Live queries replaced polling. Changes stream to agents and to enterprise UIs as they are committed, so the mesh and its dashboards stay in step without webhook chains between them.
Agent developers get a database-within-a-database: schema-optional tables they can shape freely, with native graph and vector alongside for search, relationships and memory. SurrealDB Studio sped up query design, schema evolution and debugging along the way.
Results
A simpler multi-tenant stack
Namespace isolation cut per-tenant overhead while keeping every customer's data boundary intact.
Faster iteration
Studio accelerated query design, schema evolution and debugging as the mesh grew.
A real-time experience
Push-driven updates keep the agent mesh and the dashboards that watch it in sync.
Graph and vector ready
Agent memory and knowledge are consolidated with their relationships, on the engine the mesh already runs on.
