Job hunting is often a game of volume, but automated spam risks getting you blacklisted. Enter Hermes, a new agent profile designed to work your LinkedIn pipeline from discovery to answer without going rogue. Released on November 4, 2026, this tool searches for roles on a schedule, researches employers, and tailors your resume, but it refuses to hit send without explicit human consent for external forms.

Discovery and Research Pipeline

The agent operates on 'tracks,' which are specific role definitions with their own LinkedIn queries and fit scoring. For every potential match, Hermes builds an employer dossier and decides whether to shortlist or drop the entry based on your criteria. Crucially, the resume and form answers are generated strictly from your profile facts, ensuring the AI never invents experience or skills that don't exist in your background.

Strict Guardrails and Human Control

Hermes enforces a conservative default policy to protect your LinkedIn account status. While it can prepare Easy Applications, it only sends them after you name the specific entries (e.g., 'apply 1 and 3'). External applications are never sent automatically; the agent reads and fills them using atsfill, but you must opt in to submission. Daily sends are capped at two by default, and the agent will never message or connect with anyone unless you explicitly enable those disabled tools.

Technical Setup and Risks

Running this agent requires Hermes 0.22.2 or later, a LinkedIn account, and uv for package management. It has been tested with gpt-5.6-sol via the openai-codex provider and glm-5.3 on the Z.ai coding plan. Users must be aware that LinkedIn's user agreement prohibits automated access, meaning the agent operates at your own risk through your logged-in session. If the platform flags unusual activity, your account could face restrictions or bans.

Key Takeaways

  • The agent requires Hermes 0.22.2+ and uses a pinned fork of linkedin-mcp-server for Easy Apply support.
  • Default settings cap applications at two per day to mimic human pacing and avoid detection.
  • All personal data lives in a local git repository at ~/jobsearch, separating user data from the profile code.
  • External form submissions are manual by default, requiring explicit user approval before sending.

The Bottom Line

Hermes proves that the sweet spot for AI agents isn't full autonomy, but high-friction automation that respects platform terms and user agency.