On-prem
A deployment where the agent runs entirely on infrastructure the customer controls — no agent code or customer data leaves the customer's network.
On-prem is the requirement for regulated industries and government deployments. The agent code runs in the customer's VPC or data center; the model runs against a private endpoint (Azure OpenAI, AWS Bedrock, or self-hosted).
In 2026, on-prem is mature for open-source agents (Cline, Fixie) and increasingly available for closed ones via enterprise plans. The performance gap to cloud-hosted is now small.
The real cost is operational: model upgrades, monitoring, scaling are all your problem.
Frequently asked
Can I run a closed agent on-prem?+
Sometimes. A few vendors (Sierra, Cognition/Devin) offer dedicated deployments for enterprise customers. Expect six-figure annual contracts and a multi-week setup.