How it works
How Alpacon works
Alpacon sits between AI agents, engineers, and your infrastructure. Every command passes through execution policies before reaching a server, so AI agents can operate production safely while your team stays in control.
One layer between agents and infrastructure
Three independent layers of defence
Access, Execution, and Audit work together as three independent layers, so AI agents and engineers can only run what they're authorised to.
Access that leaves nothing to attack
How agents and engineers connect. Outbound-only tunnels eliminate inbound ports entirely, and one unified identity replaces SSH key sprawl.
- Outbound-only tunnelsServers initiate encrypted connections to Alpacon. No listening ports, no exposed SSH. Attackers have nothing to scan.
- Unified identityMFA, RBAC, and centralised identity across every server. No shared credentials, no SSH key sprawl.
- Every interfaceMCP for AI agents, CLI for pipelines, API for automation, browser for humans. All governed by the same policies.
ports
Execution scoped by intent
What they can do, when. Each session declares its goal and scope, and AI validates every command at runtime against that intent, not a static allowlist.
- Intent-based session scopeA session declares its goal, server, sudo policy, and duration. No need to enumerate commands by hand. AI interprets intent from the scope.
- Dynamic AI validationCommands an agent runs are judged before they reach the host, against the purpose the session declared—so a command that looks benign on its own can still be held when it does not serve that purpose.
- Slack approval for ambiguityWhen AI isn't sure, the request routes to Slack with full context. Approvers decide in one click.
Audit powered by AI
What happened, who reviewed it. Websh and command-API sessions are recorded, session analysis maps what it finds to MITRE ATT&CK, and every command lands on one timeline you can filter.
- Full session recordingEvery terminal session, file transfer, and agent action is captured. Search and replay any session for audits or incident response.
- AI threat analysisEach session gets an AI-driven risk score. Suspicious patterns map to MITRE ATT&CK and chain detection catches attacks within a session.
- One timeline per sessionOne filterable timeline per session—every command, transfer and sudo grant, with secrets masked.
See Alpacon in your environment
A 30-minute demo with the team. We'll walk through a session on your infrastructure and answer anything technical.