Setting up a self-hosted AI agent platform like OtoDock requires checking more than just a green build. According to the GitHub documentation for OtoDock/oto-dock, the system runs agent sessions inside sandboxes on a Linux server or an Apple silicon Mac using Docker Engine 26 and Docker Compose 2.26 or later, requiring at least 4 GB of RAM and 13 GB of free disk space. When you spin up http://localhost:8400, your infrastructure security depends on correct permissions and active AI subscriptions.
Running a structured audit keeps your self-hosted infrastructure secure and prevents unvouched code or unmanaged API keys from leaking into production environments. Here is a practical ten-minute checklist you can run right on your own project setup.
1. Verify Docker and System Resources
Check that your host machine meets the hardware and runtime requirements specified in the documentation. Run your terminal and verify that Docker Engine 26 and Docker Compose 2.26 or later are active. Confirm you have allocated at least 4 GB of RAM and verified 13 GB of free disk space.
2. Inspect the Setup Wizard and Owner Account
Navigate to http://localhost:8400 or your configured server address on port 8400. Ensure the setup wizard greets you, and confirm that the primary owner account is created securely. Verify that the default Personal Assistant is installed correctly with its required tools.
3. Check AI Engine Connections
Go to the Setup and AI Engines page on your dashboard. Verify that your Claude Code or Codex subscription, API keys, or local model server (via Ollama or an OpenAI-compatible server) are connected correctly per agent, ensuring no unauthenticated engine falls back to default public endpoints.
4. Audit User Roles and Permissions
Review the platform and agent roles to ensure proper least privilege access. Verify that Admin, Manager, Creator, Editor, Member, Contributor, and Viewer seats are assigned strictly according to what team members need to see and execute.
5. Review Sharing Modes and Workspaces
Examine your agent sharing configurations. Check whether agents are set to Personal only, Personal + shared, Shared + personal, or Shared only, ensuring team members only access the workspace histories appropriate for their departments.
6. Validate Sandbox and Network Isolation
Confirm that kernel sandboxing is active for every session on the server. Verify that network isolation blocks private ranges, your LAN, and cloud metadata endpoints from the agent containers, keeping your internal infrastructure invisible to untrusted agent tool executions.
7. Test Remote Machine Pairing
If you paired a laptop, workstation, or remote PC running macOS, Linux, or Windows via a one-line install, verify that the single outbound connection is stable and that no open ports were required on the remote machine's firewall.
8. Confirm GitHub and Credential Scoping
Check that credentials stored in the PostgreSQL database are encrypted. Verify that git and gh fetch connected GitHub account tokens per command rather than storing them in plain text environment variables or files.
9. Check Scheduled Tasks and Triggers
Inspect your active background schedules, webhooks, and task triggers. Ensure that agents running tasks asynchronously notify the right channels when errors occur or when user intervention is required.
10. Audit Usage Budgets and Limits
Verify that per-user and per-agent cost budgets are configured correctly in the system to prevent runaway token consumption during long autonomous agent runs.