Skip to content

Platform Support

AI Agent Watch supports Linux, macOS, and Windows. The operating system and the AI agent determine how activity is collected and which capabilities are available.

Choose your setup path

Platform Where it can be used Setup
Linux Employee workstations, servers, containers, and Kubernetes workloads. Install Sematext Agent and enable AI Agent Watch for its Infra App. AI agents are discovered automatically.
macOS Employee workstations and other supported macOS hosts. Only Claude Code and Codex are currently supported. See configuration instructions.
Windows Employee workstations and supported Windows Server hosts. Only Claude Code and Codex are currently supported. See configuration instructions.

For supported operating-system releases, see the Sematext Agent Platform Support Policy.

Linux

The Linux collection path applies to both personal development machines and shared infrastructure. Sematext Agent automatically discovers supported AI agent processes and observes their activity through eBPF; you do not need the macOS/Windows telemetry and hook setup.

  • Activity collection: AI Agent Watch was introduced in Sematext Agent 4.5.0 and requires Linux kernel 5.8 or newer. Keep Sematext Agent up to date to collect the latest supported events and model usage.
  • Installation: Follow Linux installation for a workstation or server, or the relevant Sematext Agent deployment instructions for containers and Kubernetes.
  • Governance: Requires Sematext Agent 4.6 or later, Linux kernel 5.17 or newer, and BPF-LSM enabled. Check Governance Compatibility before relying on enforcement.

macOS and Windows

AI Agent Watch supports Claude Code and Codex on these platforms. Supported Codex use includes the CLI, codex exec, and the IDE extension. Claude Code support includes the CLI and the desktop app's Code tab.

  • Sematext Agent: Version 4.7 or later is required.
  • Collection: Configure the coding agent to report telemetry to Sematext Agent. Unlike Linux, installing the collector and enabling the Infra App alone does not configure AI-agent activity reporting.
  • Governance and additional scanning: Configure the coding agent's hooks. Governance checks known tool calls before they run; it does not enforce policy on the underlying operating-system activity.
  • Setup and verification: Follow macOS, Windows and WSL Setup for the installer command, supported coding-agent versions, managed settings, and checks.

Capability comparison

The macOS and Windows column applies to Claude Code and Codex after completing their setup. Requirements for each platform are listed above.

Capability Linux macOS and Windows
Discover AI agents automatically Yes No — configure Claude Code or Codex first.
Review agent sessions Yes Yes
See commands and tools used Yes — includes activity from processes the agent starts. Yes — tool calls reported by Claude Code or Codex. Activity inside a command is not shown separately.
See which files agents access Yes Limited — files accessed through supported tools and patches. Shell commands do not produce separate file-read events.
See network connections and traffic volume Yes No
Detect sensitive information Yes — scans captured outbound content for credentials and personal data. Yes — scans reported prompts, tool inputs, and tool results.
Block actions with governance policies Yes — blocks matching actions at the operating-system level, including actions from processes the agent starts. Requires kernel support; without it, activity is monitored but policies are not enforced. Limited — configured hooks check supported tool calls. Shell checks are best effort; MCP calls and activity inside commands are not checked. See coverage.
Track token usage and calculated costs Yes Yes

Event availability depends on the agent and version. See Captured Events for details.