Data Sovereignty

Your Data, Your Context

When you work in a context, RARS operates on a knowledge graph that contains your operational state: your business data, your conversation history, your process traces, your observations. This data needs to be protected, isolated, and under your control.

Poliglot's architecture makes a simple guarantee: your context data lives in your context. It doesn't leak to other contexts, other workspaces, or other tenants. The isolation model ensures this at multiple levels.

Serverless Isolation

Every context is provisioned as its own isolated runtime environment. This isn't a shared service processing requests from multiple tenants. Each context gets its own dedicated runtime with its own knowledge graph, its own inference model, and its own activated matrices, provisioned on demand.

Each runtime operates within its own isolated network space. Requests from one context cannot cross into another context's network boundary. There is no shared memory, no shared state, and no network path between contexts. The isolation is architectural, enforced at the network level, not just through application-level permission checks.

Checkpoints as the Storage Model

The checkpoint system is the only mechanism through which your context data is persisted. When a checkpoint is taken, the full state is captured:

• The complete knowledge graph (instance data, process traces, observations, provenance)
• Conversation history and memory state
• Operational metadata

When a context is restored, everything is rebuilt from the checkpoint. Matrices are re-activated, inferences are rebuilt, and the conversation continues where it left off.

What this means for your data:

• No data scattered across services: your context state isn't spread across databases, caches, and message queues. It's one checkpoint.
• Full restoration: a checkpoint contains everything needed to resume exactly where you left off. Nothing is lost in the restore process.

What You Control

As a workspace administrator, you control:

• What matrices are installed: each matrix you install is code that runs in your contexts. You evaluate and approve what gets installed.
• What permissions are granted: matrices request permissions, you decide what's actually allowed (see Security and Access Control).
• What data services are connected: your systems of record are your systems. You configure the endpoint URLs and authentication.

Poliglot provides the runtime. You control what runs in it and what it has access to.

Summary

• Serverless isolation: every context gets its own dedicated runtime with its own isolated network space
• No cross-context access: network boundaries prevent requests from one context reaching another
• Checkpoint-only persistence: your data lives in checkpoints, not scattered across platform services
• You control access: what matrices are installed, what permissions are granted, what services are connected

See Also

• Contexts and Agents: how contexts are provisioned and isolated
• Security: IAM policies and permission management for workspaces
• Data Governance: connecting and governing external data services
• Provenance: how every observation is tracked with full attribution