Updated 2026-10-04
Set Up Hermes Desktop with DeepSeek
Hermes Desktop is the graphical interface to Nous Research's Hermes Agent. Connect DeepSeek as the model provider to work through a chat window, choose a project folder and inspect tool activity. This guide starts with a local desktop installation and a direct DeepSeek API key.
1. Choose the package for your computer
Open the official Hermes download page and check the current platform matrix before choosing a package. The bootstrap installer, bundled desktop app and command-line installer have different requirements. Use the vendor's download link rather than a third-party mirror.
| Computer | What to check |
|---|---|
| Apple Silicon Mac | The official priority Mac platform; choose the matching desktop package |
| Intel Mac | The Hermes-Setup bootstrap DMG is ARM-only. The platform matrix separately describes darwin-x64 bundles; verify the specific release rather than using the ARM installer |
| Windows Desktop | The current MSIX package requires Windows 11 22H2 or later |
| Windows 10 CLI | The native PowerShell route has separate support; it does not imply that the current MSIX package runs on Windows 10 |
| Linux / WSL2 | Use the documented CLI route; check Desktop build instructions rather than expecting the Windows or Mac package to work |
2. Install and open Hermes Desktop
On macOS, open the matching DMG, copy Hermes.app to Applications and launch it there. On Windows, open the official .appinstaller download with Windows App Installer. If Windows rejects it, verify the package requirements and error before trying another installation method.
Bundled packages include their runtime. A package named Hermes-Setup can instead bootstrap a source installation, and Light is a remote-only variant. For the local workflow on this page, choose the package that includes a local backend.
At first launch, follow onboarding and choose the provider you intend to use. A direct DeepSeek API key does not require you to buy a Nous Portal subscription. If you postpone provider selection, configure it in Settings before sending a message.
3. Add your DeepSeek credential
Open Settings → Providers and select DeepSeek in the API-key flow. If you have multiple profiles, check Applies to before saving. Enter your DeepSeek API-platform key privately; a DeepSeek chat login is not an API credential.
For the direct provider, use its built-in routing. If the form asks for a Base URL, DeepSeek's integration guide specifies https://api.deepseek.com. Avoid defining a duplicate custom endpoint just to connect a provider already listed in the app.
A Coding Plan or third-party key must use its own delivered route. In Settings → Providers → Custom Endpoints, match the provider's Base URL and API Mode together. Do not copy a Responses or Anthropic configuration into the direct DeepSeek setup.
| Field | Direct DeepSeek setup |
|---|---|
| Provider | DeepSeek |
| Credential | API key from your DeepSeek API account |
| Base URL, if shown | https://api.deepseek.com |
| Profile | The one you will select for the test chat |
4. Set a model and check the active chat
Open Settings → Model and select the main model for that profile. Choose DeepSeek and an available model. DeepSeek's Hermes integration page uses deepseek-v4-pro as its example; confirm the exact ID is offered by your account before selecting it.
Start a new chat and check the model selector in the composer. The profile default and a particular chat's selected model are separate settings. If an older chat still shows another model, select the intended model there or start a clean test chat.
5. Verify the first reply
Paste this prompt into a new chat. Check that the message completes without a provider error and that the visible model selector still shows the intended DeepSeek model. The response text alone cannot prove which provider handled it.
A successful request uses your API account and may consume billable tokens. If the app reports authentication, balance or model errors, keep that message and follow the troubleshooting guide before submitting a larger task.
Reply with exactly: connection-ok6. Verify a small tool task
Choose a test folder containing a small text file, then send the task below. Inspect the file-tool activity and compare the reported filename and contents with the folder. A successful chat and a successful tool action are separate checks.
If the tool is unavailable, inspect tool settings and any approval request. Web search, browser use and other optional capabilities can need additional setup. Add them after this small file task succeeds.
Use the file tools to list this test folder.
Read the small text file I placed here and summarize it in one sentence.
Do not edit, delete or create files.Sources
7. Use Desktop alongside an existing CLI installation
If you already use the Hermes CLI, its documented desktop launcher is shown below. CLI and Desktop can use the same sessions, credentials and skills when they use the same data home, profile and backend. Check those settings before assuming a new window has lost your data.
Default data lives under ~/.hermes on macOS/Linux and %LOCALAPPDATA%\hermes on native Windows. WSL has its own Linux home. A Desktop window connected to a remote backend uses that machine's data; local configuration edits will not change it.
hermes desktop8. Update with the method that owns your install
Keep a backup of your Hermes data before a major update. Desktop packages use their app or package update mechanism; source installations use their documented CLI updater. Do not run an unrelated installer over the application to repair a provider error.
After updating, repeat the short chat and file test. If settings seem to have disappeared, check the selected profile and data home before changing keys. Keep the version and installation method with any error report.
Sources
FAQ
Is Hermes Desktop a DeepSeek application?
No. It is Nous Research's desktop interface for Hermes Agent. DeepSeek is the provider selected for the model requests in this guide.
Does Desktop run the DeepSeek model locally?
This setup uses the DeepSeek API. The app and tools can run on your computer while model requests are sent to the selected API provider. Local model inference is a separate configuration.
Do I have to install WSL2 on Windows?
Not for the native Windows installation. Check your chosen package: the current MSIX desktop package requires Windows 11 22H2 or later, while the CLI has a separate native Windows route.
Why does the app still use my previous model?
Check the active chat's model selector as well as the profile default. Changing the default may not change an existing chat; verify with a new chat and an explicit model selection.
Can I use a plan key instead of a direct DeepSeek key?
Use the plan's exact endpoint, model ID and protocol instructions. Do not substitute it into the direct API example unless the issuer explicitly documents that route.
Your desktop setup is ready for a larger task when the correct profile and model are selected, a short reply succeeds, and the file test shows a real tool action. Keep the working configuration as your starting point when adding tools.
Related model comparisons
Continue from this guide into structured DeepSeek-first comparison pages with model tables, routing advice, and pricing context.