Connecting more model providers to Hermes: DeepSeek, Claude, and beyond

In the last part of this series, you installed Hermes and connected a free model through Nous Portal. That was enough to get your agent thinking and calling tools. But a free starter model is a compromise: it’s fine for everyday questions and light tasks, and it quietly holds you back the moment you ask for something harder. This article is the next step in expanding Hermes deliberately — adding more model providers so your agent can take on heavier and different kinds of work.

Why add more models at all?

Your free Nous model works. So why complicate things? Because a single model forces one compromise on every task you throw at it. A model that’s cheap and fast is wonderful for summarising and routine work, but it stumbles on genuinely hard reasoning. A model that’s brilliant at analysis is overkill — and expensive — for reformatting a list. And some jobs need a capability your starter model simply doesn’t have.

Once you have a few models available, your agent stops being one-size-fits-all. You can send each job to the model that suits it:

• Cheap, high-volume work — summarising, formatting, routine lookups, the orchestration steps of an agent loop — goes to a fast, inexpensive model.
• Hard reasoning and research — careful analysis, tricky code, multi-step problem solving — goes to a stronger model, used only when it’s worth it.
• Vision tasks — reading screenshots, describing images, extracting text from a picture — need a model that can actually see, which not every model can.

This is the whole reason to expand: match the model to the task, instead of bending every task to fit one model. And it fits the pattern of this series — you’re not rebuilding your agent, you’re adding one more capability on top of what already works.

The plan: expand one provider at a time

You don’t need a dozen providers on day one. The sensible path is to add them as you feel the need, and this article follows the order that works well in practice:

1. DeepSeek first — a cheap, capable paid model that’s trivial to add and gives you an immediate step up from the free starter.
2. Claude via Command Code — when you want top-tier reasoning, reached through a single subscription instead of a separate Anthropic account.
3. Command Code as a catalog — one key that opens up dozens more models for whenever a specific job needs one.

All three are added the same way: in the Hermes Desktop settings screen. You never need the command line for any of this — if a terminal makes you nervous, you can ignore it entirely and still follow every step here.

Adding DeepSeek (the easy first step)

DeepSeek is the natural first provider to add. Its models are genuinely capable — close to flagship quality on many tasks — yet cost a fraction of the big proprietary models. In practice, DeepSeek’s fast model is an excellent everyday workhorse: cheap enough to use freely, strong enough that you rarely feel the ceiling. It’s the model I reach for most.

Here’s how to add it:

1. Create a DeepSeek account and get an API key. Sign up at the DeepSeek platform, add a small amount of credit, and generate an API key. Copy it somewhere safe — treat it like a password, because anyone who has it can spend your credit.
2. Open Hermes Desktop settings. Launch the Desktop app and go to the settings area where model providers are configured.
3. Add DeepSeek as a provider. Paste in the API key and the DeepSeek endpoint. Because DeepSeek uses an OpenAI-compatible API, Hermes talks to it the same way it talks to most providers — there’s nothing unusual to configure.
4. Select a DeepSeek model. Once the key is saved, DeepSeek’s models appear in the model picker inside Hermes. Choose the fast model for everyday work; the stronger reasoning model is there when you need it.

That’s it. You now have a cheap, capable paid model sitting alongside your free starter — and you can switch between them per task.

Two ways to reach Claude

Sooner or later you’ll want top-tier reasoning — the careful, reliable output only the very best models deliver. Claude is excellent for this, and there are two ways to give Hermes access to it. Which one fits depends on what you already have.

Route 1: you already pay for Claude (Claude Pro + Claude Code). If you have a Claude subscription and Claude Code installed on your machine, Hermes can use Claude with no API key at all — it simply calls the Claude Code command you already have, running on the subscription you already pay for. No second bill, no separate account. This is the tidiest route if you’re already a Claude user, and it’s covered step by step in its own article.

Route 2: you don’t have Claude Pro (Command Code). If you’d rather not commit to an Anthropic subscription, Command Code is a service that gives you a single API key covering a large catalog of models — Claude among them, alongside models from OpenAI, Google, DeepSeek, and many more. You paste one key into Hermes and Claude becomes selectable, no Anthropic account required. The setup mirrors the DeepSeek steps: get your Command Code API key, open Hermes Desktop settings, add it as a provider with its key and endpoint. There’s a full walkthrough — including which plan gives you API access — in a dedicated article.

The bonus with Route 2 is breadth: the same Command Code key that gets you Claude also opens the door to dozens of other models under one account. Want a specialist coding model, a huge-context model for a big document, or a vision-capable model for a screenshot task? They’re all available under the one key — you just pick them in the model list. Where DeepSeek gives you a cheap daily driver directly, Command Code gives you a whole catalog behind a single key.

Want to try a specialist coding model, a huge-context model for a big document, or a vision-capable model for a screenshot task? They’re already available under the one key you’ve added; you just pick them in the model list.

This is why the order in this article makes sense. DeepSeek gives you a cheap daily driver directly. Command Code gives you breadth — Claude plus a whole catalog — without a new account for each. Between the two, you cover almost every task an agent will ever hand you.

Picking the right model per task

With several models available, the day-to-day skill is choosing well. In Hermes you switch the active model in the interface — so you can line up the model to the job before you start:

• Routine and bulk work \u2192 your free starter or DeepSeek’s fast model. Cheap, quick, entirely good enough.
• Deep research or hard reasoning \u2192 Claude or DeepSeek’s reasoning model. Slower and pricier, but worth it when quality matters.
• Vision — images, screenshots, scanned text \u2192 a vision-capable model reached through Command Code.
• Long documents \u2192 a large-context model, again available through the Command Code catalog.

The habit to build is simple: before a task, ask which capability it actually needs, and pick the cheapest model that has it. That single question keeps both your quality and your costs where you want them.

Best practices

• Add providers as you need them. There’s no prize for wiring up ten providers at once. Start with DeepSeek, add Command Code when you want breadth, and stop there until a real need appears.
• Guard your API keys. Each key can spend real money. Keep them out of screenshots, chat logs, and shared files, and rotate any key you suspect has leaked.
• Watch your spend early. Every provider has a usage dashboard. Check it in the first few days so a runaway agent loop can’t quietly rack up a bill.
• Keep the free model around. It’s still the right choice for genuinely trivial work — no reason to pay for what it already does well.

Frequently asked questions

Do I have to remove the free Nous model to add others?

No. Providers sit side by side in Hermes. Adding DeepSeek or Command Code doesn’t replace your free model — it just gives you more options in the model picker. Keep the free one for light work.

Why not just add a direct Anthropic key for Claude?

You can, and it works. But a direct key means another account and another bill to manage. Reaching Claude through Command Code folds it into one subscription that also covers many other models, which is simpler if you expect to use more than just Claude.

Which model should I use for reading images?

You need a vision-capable model — not every model can process images. The simplest route is to pick one from the Command Code catalog when you have an image task, then switch back to a cheaper text model for everything else.

Can I really do all of this without the command line?

Yes. Adding providers, pasting keys, and switching models all happen in the Hermes Desktop settings and interface. The command-line tools exist for people who prefer them, but nothing in this article requires them.

What’s next

Your agent is no longer tied to a single model. You’ve added a cheap, capable daily driver in DeepSeek, a route to top-tier Claude through Command Code, and a whole catalog behind one key for the occasional specialist job — and you can match any task to the model that suits it. That’s another deliberate step in expanding Hermes: each part of this series leaves you with a more capable assistant than the last.

Next up: if you already have Claude Code installed on your machine, there’s a second way to reach Claude — without any API key at all. The next article shows how Hermes calls the local Claude Code command directly, why it works on plain Claude Pro, and the one Windows gotcha to know before you try it. This article is part of the Agentic AI series.