All posts

Personal Agents: Hermes Agent and Microsoft Scout

22 October 2026 · NavySteel

OpenClaw is no longer the only always-on personal agent you can run. If you want the convenience without OpenClaw's open defaults — or you live in the Microsoft 365 world — two others are worth knowing. (For OpenClaw itself, see our companion post, OpenClaw with Claude on macOS and Windows.)

Microsoft Scout launched at Build 2026 as an always-on personal agent for Microsoft 365, and it is built on OpenClaw. It works across Teams, Outlook, OneDrive and SharePoint. A desktop app adds browser activity, local resources and MCP servers. Each agent runs under its own governed Entra identity, with Purview policies enforced inline. Access is limited to Microsoft's Frontier programme for now.

Hermes Agent from Nous Research is an MIT-licensed personal agent built to run unattended and build up knowledge over time. It writes reusable skills after complex tasks. It reaches you on more than 20 platforms through one gateway, supports MCP and offers five sandbox backends. Hermes Desktop has been a native app for macOS 12+ (Apple Silicon only) and Windows since June 2026. It is still a public preview.

Feature comparison

Feature Microsoft Scout OpenClaw Hermes Agent
Built on OpenClaw — Own codebase
Focus Microsoft 365 work Personal assistant, device control Personal assistant, self-improving
Always-on and proactive Yes Yes, heartbeat and cron Yes, scheduled tasks
Learns your habits Yes Yes, memory Yes, writes its own skills
Chat channels Teams ~40, including native iMessage 20+, including Teams
Mac device control Browser and local files Native menu-bar app: camera, screen, shell Through tools only
MCP Yes Yes Yes
Sub-agents — Yes Yes
Sandboxing Microsoft-managed Docker, Podman, SSH (opt-in) Five backends
Identity and governance Entra identity per agent, Purview Config file and audit command Config-based
Mac app Yes Stable Public preview, Apple Silicon only
Main risk Low, enterprise-governed Skill marketplace, open defaults Newer, less battle-tested

Can you use a subscription instead of an API key?

API billing adds up quickly for an always-on agent. Whether you can use a flat monthly subscription depends on the model provider, not the harness.

Harness Claude Pro or Max ChatGPT Plus or Pro
OpenClaw No, blocked since April 2026 Yes, through Codex sign-in
Hermes Agent No, blocked since April 2026 Yes, through Codex sign-in
Microsoft Scout No, uses Microsoft's models No, uses Microsoft's models
  • OpenClaw: subscription sign-in and API-key billing are kept separate. When both are set up, OpenClaw prefers the subscription. If you hit a Codex usage limit, it pauses that sign-in until the reset time.
  • Hermes Agent: you sign in with your ChatGPT account through a device-code flow, with no API key. Hermes can also hand execution to Codex as its engine, paid for by the same subscription. OpenAI does not document how this usage counts against your plan, so watch your limits early on.
  • Claude Max API proxy: OpenClaw lists one as a provider. Avoid it. Anthropic blocked third-party subscription use, so routing your plan through a proxy puts your account at risk.
  • Claude on a subscription: stay first-party. That means the Claude app, Claude Code Channels for Telegram and Discord, or Claude Tag in Slack.

Which to pick

  1. Closest to Scout for personal use at a flat cost: Hermes Agent with a ChatGPT subscription through Codex sign-in. Run it under a separate Mac user or a VM, with a sandbox backend on.
  2. Deep Mac control and iMessage: OpenClaw with the same sign-in and the hardened baseline above.
  3. Claude on your existing plan: Anthropic's own apps and integrations.

If you use Scout at work, keep any personal agent fully separate. Use no work accounts, no corporate device and no company tenant connectors. A personal agent has none of Scout's identity and compliance controls, and mixing the two will likely breach your employer's policies.

Was this helpful?