​
Managed agents on BYOC

A managed agent is an agent definition that is authored in the Allternit platform but executed on a runtime you control. This gives you custody of execution while keeping the authoring experience central.

​
When to use managed agents

  • You need execution to stay inside your network or region.
  • You want to use custom models or inference endpoints.
  • You want to reuse existing compute rather than pay per-request platform markup.

​
Architecture

  1. Author the agent in the dashboard or via the Agents API.
  2. Link a runtime with the agent-daemon and a BYOC deployment.
  3. The platform dispatches runs to your daemon; the daemon executes them locally.

​
Deployment steps

​
1. Provision a runtime

Install the agent daemon on a host that can reach your models and tools:

​
2. Create a BYOC deployment

​
3. Bind agents to the deployment

Set deployment_id when creating or updating an agent. All runs for that agent are routed to the linked runtime.

​
4. Monitor runs

Use the dashboard or GET /v1/deployments/{id}/runs to view run history, logs, and health for your managed agents.

​
Security

  • The daemon authenticates with mTLS and refreshes short-lived credentials automatically.
  • Secrets are stored in Vault and injected at runtime.
  • Audit logs flow back to the platform for compliance reporting.

​
Scaling

Run multiple daemons behind a load balancer and register them against the same deployment. The platform distributes work across healthy daemons using the work queue.