Deployment Models
How the Company Brain can run: fully managed cloud, isolated single-tenant, and sandboxed or self-hosted footprints for regulated environments. Understand the options and what each one means for data residency and control.
The Company Brain is designed to meet companies where their data-residency and control requirements already are. The platform separates a managed control plane from a deployment footprint you choose, which is what makes the range of models below possible.
Deployment options#
The fastest path. Parcle runs the full platform for you as a managed service. Your company's brain is logically isolated and governed by your access policies. No infrastructure to operate.
A dedicated, single-tenant instance operated by Parcle on your own host. Suited to companies that need physical separation from other tenants and dedicated capacity.
The platform runs inside your own boundary, either in a sandboxed environment or on your infrastructure, so data and model traffic stay within your network perimeter. Suited to regulated industries and air-gapped or "sandboxed AI employee" patterns.
Choosing a model#
| Requirement | Managed Cloud | Isolated Single-Tenant | Self-Hosted |
|---|---|---|---|
| Time to first value | Fastest | Fast | Scoped during onboarding |
| Tenant separation | Logical | Dedicated instance | Within your boundary |
| Data residency control | Region selection | Dedicated region | Full control |
| Runs in your network perimeter | No | Optional | Yes |
| Bring-your-own / local models | N/A | Supported | Supported |
| Operational responsibility | Parcle | Parcle | Shared / you |
The right model is scoped per engagement. Data-residency, key-management, and model-routing specifics are confirmed during onboarding rather than fixed defaults. This page describes the shapes available, not a one-size-fits-all configuration.
Bring-your-own and local models#
Because the reasoning layer is decoupled from any single model provider, the Company Brain can route to:
- Hosted frontier models for the managed and single-tenant models.
- Open-source / local models running inside your environment, appropriate for sandboxed deployments where no traffic may leave the perimeter. This is the pattern used when the Company Brain backs a self-contained, sandboxed agent that never reaches the public internet.
See Integrations for how agents connect to the brain regardless of where the model runs.
What stays consistent across models#
Whatever the deployment, the contract to your agents is identical:
- The same way to read from and write to the brain.
- The same company-wide governance: permissions, roles, SSO, and audit. See Enterprise.
This means you can start on managed cloud and migrate to an isolated or self-hosted model later without changing how your agents connect.
How data sources feed the brain — connect once, available to every permitted agent.