One connection layer
Build on Nexum MCP rather than deploying another management backend.
AI-NATIVE SERVER OPERATIONS
Product preview · Not yet released
Your servers.Managed from ChatGPT.
Inspect Linux hosts, plan Docker Compose deployments and diagnose running services in natural language. Nexum handles the connection and authorization; you decide the scope.
- Nexum MCP
- OAuth Agent grants
- Native Docker Compose
Inspect the services on prod-app-01 and assess the risks of a Compose update.
- 01project.listChoose a unique currently visible target
- 02project.openPorts, network, images and persistent data
- 03files.* / process.*Stop if authorization or prerequisites are missing
Example command · Not run by this page
docker compose config --quietThis site illustrates the expected workflow; it is not a remote console.
THE IDEA
Not another PaaS. An operations interface for AI.
Keep your services where they are. You do not need another deployment control plane. Steward reuses Nexum Cloud, OAuth grants, Agents and the existing MCP Project / Files / Process capabilities.
Explicit machine scope
Humans select Agents by name; Nexum enforces stable ID authorization.
Native Compose
Use Docker Compose instead of inventing another orchestrator.
PRODUCT DIRECTION
Designed for operational work, not tool lists.
These are planned capabilities, not yet validated end to end on a live server.
Server inspection
Inspect CPU, memory, storage and services from real host state, not assumptions.
Compose deployment
Review stacks, configuration, ports, images and volumes before authorized changes.
Logs and diagnosis
Use container state and logs to explain failures with verifiable evidence.
Service maintenance
Ground updates and recovery in actual versions, health checks and data restoration prerequisites.
HOW IT WORKS
Four steps from access to action.
- 01
Install Nexum
Set up Nexum in your own account.
- 02
Connect server Agents
Install a Nexum Agent on each Linux host you want to manage.
- 03
Choose authorized machines
Sign in using Nexum OAuth and explicitly select the allowed Agents.
- 04
Operate from ChatGPT
Reuse Nexum MCP to inspect targets, assess risk, execute within scope and verify.
TRUST BY DESIGN
Automation needs boundaries.
The ability to run a command is not the same as safe, autonomous server operations.
Names are not credentials
Agent names identify machines; backend grants and stable Agent IDs enforce authorization.
Shell is not a sandbox
process.exec has host-user privileges. Docker sockets, databases and volumes need separate protection.
Success needs evidence
A completed command is not a healthy service, and rolling back an image does not restore data.
OPEN DEVELOPMENT
Built in the open. Verified in stages.
The Nexum MCP package and website source are public. ChatGPT installation, product Skills and real-server operations still require later integration and validation.
CURRENT STATUS / DESIGN PREVIEWNever post server credentials, tokens or production logs in public issues.