Hermes Agent Is INSANE (FREE!)
This video addresses common FAQs about Hermes, a free open-source AI agent, covering how to manage multiple tasks, how it compares to Claude Co-work, and strategies for reducing API token costs. The presenter demonstrates features like the Kanban board, multiple agent profiles, memory systems via Obsidian, and OAuth-based model integrations. The content is aimed at users of the 'AR Profit Boardroom' community.
Summary
The video opens by showcasing Hermes agent's capabilities, including voice agents, image and video generation, text-to-speech, and an autonomous 'goals mode.' The presenter then works through a series of FAQs submitted by members of the AR Profit Boardroom community.
On the topic of handling multiple tasks simultaneously, the presenter explains two main approaches: using the Kanban board (referred to as the 'camb board') to break a large task into subtasks managed by sub-agents, and creating multiple Hermes profiles via the terminal command 'hermes profile create' to run entirely separate agents in parallel. The presenter notes that most users will not need the multi-profile approach.
For organizing long-term projects without interference, the presenter recommends creating reusable 'skills' for recurring workflows and setting up a memory system using Obsidian. Obsidian stores project context in a knowledge graph, reduces token waste from re-explaining context, and is compatible with other AI tools like OpenClaw and Claude.
When comparing Hermes to Claude Co-work, the presenter argues that Claude Co-work is better suited for non-technical beginners focused on text-based knowledge work, while Hermes is more customizable, open-source, and capable of multimodal tasks like image and video generation. The presenter notes they use Claude Code heavily for large builds but prefers combining both tools in a unified 'mission control.'
On the question of whether Hermes can replace tools like Screaming Frog or Ahrefs, the presenter draws a clear line: if a tool provides proprietary data unavailable elsewhere (e.g., Ahrefs backlink data), it cannot be replaced. If the tool is used for building or organizing, Hermes can substitute it.
For managing token and API costs, the presenter outlines three strategies: using a persistent memory system to avoid re-prompting, leveraging free APIs such as 'Step 3.7 Flash' available via Newsportal for 30 days, and using OAuth logins (e.g., via Twitter/Grok, MiniMax M3, or ChatGPT subscriptions) to avoid additional API fees entirely. The presenter also notes that Hermes becomes more token-efficient over time with use, and that the development team actively releases updates to reduce token consumption.
Key Insights
- The presenter argues that Hermes' biggest differentiator from Claude Co-work is not task performance but rather that it is open-source, free, and allows users to plug in their own APIs — including non-Anthropic models — whereas Claude Co-work is locked into Claude's ecosystem.
- The presenter claims that a significant portion of token costs for many users comes from reexplaining the same context repeatedly, and that integrating a persistent memory system like Obsidian can eliminate that waste.
- The presenter draws a clear boundary on AI tool replacement: if a tool's value is proprietary data (like Ahrefs' backlink and traffic estimates), AI agents cannot replace it, but if the tool's value is in execution or organization, Hermes can substitute it.
- The presenter demonstrates that OAuth logins — such as using an existing Twitter subscription to access Grok within Hermes — can entirely eliminate the need to pay for a separate API, effectively unlocking model capabilities at no additional cost.
- The presenter states that MiniMax M3's coding plan, accessed via OAuth inside Hermes, unlocks multimodal capabilities including image generation, video generation, voice notes, and real-time voice conversation — features unavailable in tools like Claude Co-work.
Topics
Full transcript available for MurmurCast members
Sign Up to Access