Video: "Hermes Agent + DeepSeek V4 Free = The Most Powerful AI Workflow You're Not Using Yet" by Julian Goldie on YouTube.
Why DeepSeek V4 Flash specifically
DeepSeek V4 Flash is a model designed with agent workflows in mind — it handles tool calls, multi-step reasoning chains, and sequential task execution more reliably than many general-purpose models at the same tier. The context window is large enough to hold a full task brief plus previous session memory without hitting limits mid-run.
It being free on the Nous Portal changes the calculation for small teams. Previously you chose between Claude (excellent but costs add up), a cheap OpenRouter model (inconsistent on agent tasks), or running something locally (viable but requires hardware and setup time). DeepSeek V4 Flash on the Nous Portal is a fourth option: cloud-hosted, agent-optimised, and zero cost at the access tier Julian used in the video.
What the setup actually involves
You point Hermes at the Nous Portal API endpoint and set your model to DeepSeek V4 Flash. That is roughly a five-minute config change if Hermes is already installed. The Nous Portal gives you an API key without a billing setup, and the free tier covers enough usage for personal or small-team automation work.
Once running, Hermes operates the same way it does on any other model: it takes a goal, breaks it into tasks, executes them using whatever skills you have installed, and reports back. The difference is that nothing leaves your wallet.
What it handles well
Julian ran the combination through content research tasks, web scraping jobs, and a SEO keyword clustering workflow. The results for structured, repeatable work were comparable to paid model outputs. Hermes + DeepSeek V4 Flash is well-suited for the kind of automation that runs overnight: pulling data, organising it, generating structured documents, sending summaries. The "set it before you sleep, check it in the morning" use case.
The persistent memory that Hermes maintains across sessions also carries over properly with this model — your agent remembers client context, past results, and workflow preferences without you re-briefing it every run.
Where a paid model still wins
For tasks requiring editorial judgement — writing that needs to sound distinctly human, nuanced client communication, anything where tone matters — Claude Sonnet or Opus still produces noticeably better output. DeepSeek V4 Flash is consistent and capable, but it is not a substitute for the models at the top of the reasoning tier.
There is also a rate limit on the free tier. Heavy parallel workloads or swarm-mode runs with multiple simultaneous agents will hit it. For those use cases you either need a paid model or a longer runtime that spreads the work out.
Where this connects to NordSys
We build and maintain automation systems for clients — which often means choosing the right model for each part of the pipeline rather than defaulting to one tool for everything. DeepSeek V4 Flash on the Nous Portal is a genuinely useful free option for the structured, repetitive work that makes up most AI automation. We can configure this alongside Claude for the work that needs it.
See our Programming & Automation service →