Hermes + Claude Code: shared memory and workflow setup

The speaker pairs Claude Code as a coding assistant with Hermes as an operator that runs scheduled jobs on a server or personal machine. His examples cover website signup monitoring, feedback on video editing and client operations. These are accounts of his own workflow, rather than independent evidence of reliability.

Shared memory connects the two tools. The tutorial describes pointing both at the same memory account and identity, with an added layer that preserves their existing memory systems. The speaker uses Mem0 and briefly names Zep, Letta and Supermemory as alternatives. This overview uses an account, identity and API key; it does not document provider-specific configuration or a fully offline shared-memory setup. His demonstration stores food preferences through Claude and then asks Hermes to recall them.

For routine operation, he recommends a cheaper model or a local LLM, reserving a stronger model for building. He also advises designing the handoff before development, requiring each job to verify its result and asking for approval before sending email, spending money or deleting data. Hermes can call Claude Code when a task needs new code; Codex is another suggested builder.

The speaker says corrections can become reusable skills, but acknowledges that his workflows needed fixes after launch. His practical advice is to start with one working job before adding more.