The problem
Remote workflows require working authentication, network access and session handling. Before using an agent with ssh user@server, check these requirements:
- Interactive authentication — the client needs a way to handle the prompts your server requires
- SSH keys on your machine — giving an AI agent access to your private keys is a security nightmare
- Jump hosts — verify network access, credentials and session handling for each hop
- Remote interactive tools — once SSH'd in, you need to run htop, vim, tail logs — all interactive
The result: developers SSH in themselves, do the work, then copy-paste findings back to the agent. The agent loses all context of the remote environment.
How agend solves it
Your agent operates from its own isolated Linux machine in the cloud. From there, it can SSH into your infrastructure just like a human operator would:
- Password auth works — real PTY means SSH password prompts are handled natively via
shell_send_raw - SSH keys stay in agend — store deploy keys in the agend environment, not on your laptop. Use credentials scoped to the test workflow.
- Jump hosts / bastion servers — the agent SSHs to the bastion, then SSHs to the target. Nested interactive sessions just work.
- Remote tools work — once connected, the agent can run htop, tail logs, edit configs, restart services on the remote machine
- Session persistence — the SSH session stays alive. The agent can do other things and come back to it.
Security model
agend adds a security layer that running SSH from your laptop doesn't have:
- Isolated origin — SSH connections come from the agend environment, not your workstation. External systems remain reachable with the credentials and permissions you grant.
- Separate credentials — deploy keys and service accounts live in agend, not in your personal
~/.ssh - Workflow records — retain the transcripts and verification evidence your client provides; confirm they meet your requirements
- Disposable — if something goes wrong, destroy the environment. Your machine is unaffected.
What you can automate
- Production incident investigation across multiple servers
- Configuration drift detection and remediation
- Log aggregation from servers without centralized logging
- Multi-server deployment verification
- Database access through SSH tunnels
- Network equipment management (routers, switches with CLI)