A checklist of the decisions and confirmations a self-hosted deployment needs, with who owns each one.
Infrastructure and network
| Item | Owner |
|---|
| AWS Region, outbound path (NAT or proxy), TLS from the load balancer to ingress, network location of the model gateway, and what lies inside the private upstream ranges you give the egress proxy | Your team |
| Encryption at rest, keys, backups and retention for Amazon RDS, Amazon S3, cluster volumes and EKS secrets; trace retention | Your team |
| Network policy enforcement and address-spoofing protection in the cluster (S-14) | Your team |
| The default ServiceAccount role (S-12) and IAM wildcard actions (S-13) | Both |
| Use and approval of Composio (S-16); whether a media generation provider key is configured | Both |
| Encryption in transit between components inside the cluster | Context |
| Application replica count; live inbound sandbox ports; Box transfer link lifetime | Context |
Agents and data
| Item | Owner |
|---|
| Agent policy presets in use, and any allow, ask or deny rules (see Agents and approvals) | Context |
| Whether logs contain prompt text and tool output; scope of ClickHouse data; retention of the run event stream | Context |
| Organization settings: who may create agents and workspaces; whether all workspaces are visible to admins | Both |
| Tests: running another person’s assistant (P-01) and an organization admin’s assistant (P-02) | Context |
| Runtime test: whether private-drive metadata is visible through file identifiers | Context |
Accounts, audit and support
| Item | Owner |
|---|
| Sign-in settings: domain mapping, fallback organization, default sign-up workspace, email verification, password sign-in, single sign-on (S-11) | Both |
| Platform superadmin accounts and holders of the setup password | Both |
| Retention of authorization, approval and audit records, and their export to your security tooling | Context |
| Context’s access to the environment for support (identities, approval, session records) and data sent to Replicated | Context |
| Data deletion when users leave and at contract end; incident response and notification | Both |
Checklist
- Choose the sandbox egress posture and allowlist (see Sandbox egress).
- Confirm network policy enforcement in your cluster.
- Restrict sign-up to your email domains, or use single sign-on only.
- Decide whether the organization infrastructure page stays off.
- Choose agent presets and the agent and workspace restrictions in Platform settings.
- Limit connector targets to your MCP servers and turn on the connector network policy.
- Decide load balancer source ranges and a WAF.
- Agree audit export, retention and data deletion with Context.
