← Back to blog

GPT-5.6, HY3, Clacky: Hermes Agent roundup, July 10

hermesgpt-5.6hy3clackyroundup
GPT-5.6, HY3, Clacky: Hermes Agent roundup, July 10

GPT-5.6 is live in Hermes Agent

@NousResearch announced GPT-5.6 support in Hermes Agent, available directly through Nous Portal:

The announcement hit 1,961 likes, 96 retweets, and over 129,000 impressions. Community response was fast - @mr_r0b0t noted his Hermes Agent is "eating good this month," and @Teknium replied to Sam Altman's GPT-5.6 post with "Makes me happy to not have to give anthropic more money."

HY3 free in Hermes via Nous Portal

@JulianGoldieSEO highlighted that Tencent's HY3 (295B model) is now free inside Hermes Agent through Nous Portal:

Setup is two terminal commands - run hermes model, select Nous Portal, pick HY3. The model is free for two weeks. Julian also published an OmniRoute integration that gives Hermes "a free coding brain" with auto-fallback, compression, and provider switching through one local gateway. His MoA comparison showing Hermes Mixture of Agents outscoring Fable 5 on a 47-task benchmark also picked up reach.

Clacky adds Hermes background agents

@WuyepRaynan updated Clacky, his open-source Windows AI companion, with Hermes Agent-powered background agents:

Clacky's parent announcement (675 likes, 105 retweets, 37K impressions) positions it as a Windows-native alternative to Clicky. The Hermes integration adds background agent execution with Composio for app connectivity across 1,000+ tools.

Hermes Agent Masterclass finale

@tonbistudio published the final module of his Hermes Agent Masterclass series, covering security:

The video covers approval gates, permission scoping, and Hermes's built-in security controls for multi-agent setups.

Community workflow tips

@MGT_maccha shared a read-only /goal prompt pattern for inventory management - it lists memory, skills, and cron entries, then runs classification and prioritisation without modification.

@Suraj_sharma14 published a guide on structuring Hermes agents as a team of specialists - one for research, one for content, one for SEO. His approach mirrors the pattern of specialized sub-agents seen in production Hermes deployments.

@SmelterLabsai logged 5 substantial commits in the last 24 hours, including a cleaner Hermes Cloud connection mode - sign in once, discover your agents, connect silently. Less setup friction for desktop users moving between machines.

@0xOzp shared his DeepSeek API usage data - 1 billion tokens at $14.58 over 30 days, running deepseek v4 pro and v4 flash through Claude Code and Hermes Agent. A data point on the cost efficiency of mixing provider tiers with Hermes's model routing.

Termagotchi
_

Ryan Underdown

Autodidact. Rarely listens to advice.

Follow on X @catamarammed or GitHub @underdown