Hermes Agent v0.19 Quicksilver: 80% Faster, Durable Jobs, Agent Fleets

Key Takeaways

• ~80% faster cold start. Time from request to first token drops from 4.3s to 0.9s, and the agent streams its reasoning in real time instead of showing a blank screen.
• Crash-durable jobs. Background jobs and final replies are persisted to a database, so if the app or gateway crashes mid-task, the work finishes and delivers when it comes back.
• One gateway, a fleet of profiles. A single Hermes gateway can route each channel to its own specialist profile (e.g. SEO, writing, ops), and the main agent auto-delegates to sub-agents — each profile can even use a different model or API.
• Smarter approvals with a safety default. Choose manual approval or full-auto, but the new default is a second, independent AI that assesses each flagged command before it runs. Deny rules you write hold even in full-auto mode.
• Secrets, sessions, and a panic button. Real secrets management keeps keys out of plain-text config; sessions export to Markdown/HTML with a redact flag that scrubs secrets first; and a new safe mode boots Hermes with every customization off to recover from a broken config.
• Two new free models (referenced as HY3 and Laguna S 2.1) can be assigned to different profiles so tasks auto-route around rate limits. The release reportedly landed 2,245 commits at once.

Commands & Features Mentioned

# Update from the Hermes manage/OS panel: click "Update Hermes"
/subscription        # view billing without opening the dashboard
# Export a session to Markdown / HTML, with a --redact flag to scrub secrets
# Safe mode: boot Hermes with all customizations off to fix a broken config
# Per-profile delegation via the built-in delegate tool

Note: this recap ends with a promotion for the creator's paid community. The feature details above reflect the v0.19 release notes as presented; verify specifics against the official Hermes changelog before relying on them.