Launch an AI agent on your computer - in 15 minutes with Hermes Agent
Install the open-source Nous Research agent, connect DeepSeek with your own key, and test it with a first task - no cloud panels and no sharing keys with third parties.
- 1.Install Hermes Agent - the open-source agent from Nous Research that runs on your computer. macOS, Linux, or WSL2: run curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash in the terminal. Windows: run iex (irm https://hermes-agent.nousresearch.com/install.ps1) in PowerShell.
The only requirement is having Git installed. If you are not used to the terminal, the website offers regular installers for macOS and Windows (DMG and EXE).
Hermes Agent ↗ - 2.Restart the terminal and run the hermes setup command - the setup wizard will open. Select Quick Setup: it will ask for the model provider, key, and address.
If you don't want to create API keys, there is a way without them: hermes setup --portal - sign in via your Nous Portal account using OAuth. The Portal has a free tier, while paid ones give monthly credits and access to hundreds of models.
Hermes Agent ↗ - 3.Get your DeepSeek API key on the DeepSeek platform (API keys section) and enter it in the wizard: provider - DeepSeek, Base URL - https://api.deepseek.com, model - deepseek-v4-pro. These are the values from the official DeepSeek guide for connecting to Hermes.
An API key grants access to your paid account: do not paste it on third-party sites or put it in a repository. The model for Hermes must support a context of at least 64,000 tokens - deepseek-v4-pro fits the bill.
DeepSeek V4 Pro ↗About this tool - 4.If you need a second provider - like Kimi - open the AI Providers documentation section: it lists supported providers (including kimi-coding and kimi-coding-cn) and shows how to set up a fallback chain of models in the Hermes configuration.
The idea behind the fallback chain is simple: the main provider handles most tasks, and the backup takes over when the main one is unavailable or hits a limit.
Kimi K3 ↗About this tool - 5.Launch the agent with the hermes --tui command (terminal interface) or just hermes. Give it the first task - copy the prompt below, replacing the data in the square brackets.
Run the agent from your project folder: it works with the files in the directory from which it was called. The official recommendation is to start with a task whose result is easy to verify.
Hermes Agent ↗PromptRead the files in this folder and produce a brief for me. Steps: 1) List the files you actually read. 2) Summarize what changed since [DATE] in no more than 10 bullet points. 3) End with an «Action items» section: at most 5 concrete next steps, each starting with a verb. Rules: use only facts from the files, never invent numbers or names; if something is unclear, say so explicitly instead of guessing.
What this prompt doesThe prompt is in English - the agent understands it more accurately. Replace [DATE] with the date from which to count changes. The agent will read the files in the current folder, compile a summary of no more than 10 points, and finish with a list of concrete actions. The first line of the response is a list of read files: it immediately shows whether the agent reached your data or not. - 6.Check that the session is saved: close the agent and run hermes --continue - it should resume the previous conversation. If continuation works - the setup is complete.
This check is described in the official quick start: until sessions are saved, it is too early to move on to more complex scenarios.
Hermes Agent ↗
- freeHermes AgentThe agent itself is open-source and free, installed on your computer - you only pay for the model you connect.
- APIDeepSeek APIPay-as-you-go. Check the current price per million tokens on the DeepSeek pricing page - it can change.
- subscriptionNous PortalA path without your own keys: there is a free tier, while paid ones provide monthly credits and access to hundreds of models.
Why today
AI services change fast - interfaces and free limits may differ from what's described.