Playing this video loads it from YouTube (Google), which may store cookies on your device and receive your IP address. Click to play.
Video: "DeepSeek V4.1 Flash + Hermes Agent (FREE!)" by Julian Goldie on YouTube.
What the three-way connection actually is
The setup combines three components. Hermes Agent is an open-source AI agent framework from Nous Research - it handles scheduling, tool use, memory management, and sub-agent coordination. DeepSeek V4.1 Flash is the inference model powering the agent's reasoning - released on 10 September 2026, it is smaller and cheaper than the previous V4 Pro but outperforms it on most benchmarks. Token Harbour is an API broker that routes requests to various AI models, including DeepSeek V4.1, with a free tier that covers everyday use without requiring a paid API key.
In practice, Hermes talks to DeepSeek via Token Harbour as though it were any other API endpoint. The agent gains access to a capable, up-to-date model without a monthly bill. That is the cost side of the equation. The memory side is handled separately by Obsidian.
What Obsidian memory adds to an AI agent
Most AI agents have no memory between sessions. Each conversation starts from scratch: the agent does not know what it concluded last time, what files it created, or what tasks it left in progress. Token limits make it impractical to paste in the full history with every new request.
The Obsidian integration solves this by writing agent outputs - task summaries, decisions, research notes, completed actions - into a local Obsidian vault. When the agent starts the next session, it reads the relevant notes first. The result is something close to persistent context: the agent can pick up where it left off, avoid repeating itself, and build up a genuine working memory over time. For repetitive workflows - weekly reporting, ongoing research, client account tasks - this makes a material difference to how useful the agent actually is.
What tasks the stack handles well
The free setup is well-suited to tasks that run regularly and benefit from remembered state. SEO monitoring is an obvious fit: the agent researches keywords, logs findings to Obsidian, and on the next run already knows what has been covered. Content workflows work similarly - draft, review, and refine cycles can accumulate context without the agent forgetting the brief. Internal process automation - triaging emails, summarising documents, updating a task list - stays manageable within Token Harbour's free tier as long as token volumes stay moderate.
To be fair, there are limits. Token Harbour's free tier has daily caps, so high-volume or time-sensitive workloads will hit them. DeepSeek V4.1 Flash is strong on reasoning and coding tasks but not a specialist in every domain. And maintaining the Obsidian vault - keeping it structured so the agent can actually read it usefully - requires some upfront setup work that the video does not cover in full.
Why the cost angle matters for smaller operations
The AI agent market is largely framed around enterprise budgets. Managed agent platforms from major providers charge several hundred pounds a month for capabilities that, in practice, a carefully configured free stack can approximate. That does not mean the enterprise tools are overpriced - they add reliability, support, and integrations that this setup does not include. But for a small team that wants to experiment with agentic workflows before committing to a platform, starting with a free open-source stack is a reasonable first step.
Worth knowing: DeepSeek V4.1 Flash's weights are MIT-licensed, which means you can run the model locally as well as via Token Harbour. If you have suitable hardware (or a modest VPS), you can eliminate the API dependency entirely and run the whole stack without any usage limits.
Where this connects to NordSys
Setting up Hermes, Token Harbour, and an Obsidian memory layer is a technical task - configuration is not complicated, but getting the prompts right and structuring the memory vault to be actually useful takes time. Our AI Agents service covers exactly this kind of setup: starting with an assessment of what you actually want the agent to do, through to a configured, tested system that fits your workflows from day one.
See our AI Agents service →