Agent Teams

What Agent Teams does

The README casts each Hermes instance as one worker and Agent Teams as the layer that coordinates several of them around the clock on a machine or VPS. A dashboard shows each agent's state, live execution stream, message history and cost. Supervisor agents review output and nudge drift, loop and stall detection is built in, and per-team daily budgets pause spending automatically.

SQLite task queues recover in-flight tasks after a restart, and context compaction, tool-result aging and stale-payload stubbing keep sessions bounded. A human inbox lets agents ask a question and resume once you answer, and a browser takeover feature using CDP screencast lets you handle CAPTCHAs and 2FA. An AI team builder called The Architect designs teams, and the README gives marketing and sales teams as examples. It lists 40+ providers through Hermes, or a single OpenAI-compatible endpoint.

Key features

• Dashboard with each agent's state, live stream, messages and cost
• Supervisor agents and automatic loop and stall detection
• Per-team daily budgets that pause spending
• Human inbox for questions and credentials, plus browser takeover for CAPTCHAs and 2FA
• SQLite task queues that recover in-flight work after a restart
• The Architect, an AI team builder

When to use it

• Run a marketing or sales agent team overnight and read a summary in the morning
• Cap agent spending with daily budgets per team
• Hand a credential or a CAPTCHA to an agent through the human inbox

Who it is for: Hermes Agent users who want several agents working together unattended, with oversight and cost limits.

How it fits with Hermes Agent

Built around Hermes Agent: Hermes instances are the workers, and Agent Teams adds supervision, scheduling and cost control on top.

FAQ