Hermes Agent + VS Code: Complete Editor Integration Guide

{
  "acp.agents": {
    "Hermes Agent": {
      "command": "hermes",
      "args": ["acp"]
    }
  }
}

hermes acp --check

Optional Browser Tools

If you need Hermes to interact with web pages (e.g., for testing or scraping), you can enable browser tools:

hermes acp --setup-browser

This command installs the necessary browser dependencies. After running it, reconnect the VS Code ACP Client. Browser tools allow Hermes to navigate to URLs, take screenshots, and extract content. This is useful for debugging web applications or verifying deployments.

When to Use VS Code vs. Other Surfaces

The official documentation is clear about the different use cases:

• Use Copilot for fast inline suggestions and editor native completions. Copilot excels at single line or small block completions as you type.
• Use Hermes in VS Code for permissioned agent work inside the repo. This includes multi file refactors, debugging loops that need terminal output and source edits, and repo specific workflows where Hermes should remember conventions through skills and memory.
• Use FlyHermes when you want the hosted cloud version instead of maintaining the local server, gateway, provider credentials, and dashboard yourself. FlyHermes is the cloud path for managed Hermes.
• Use GitHub integration for unattended PR, issue, CI, and webhook work when the editor is closed.

Running Hermes on a VPS with Telegram (The ZazenCodes Workflow)

A significant portion of the source material covers the workflow demonstrated by ZazenCodes, where Hermes runs on a Hostinger VPS inside a Docker container, and you interact with it via Telegram from your phone. This is not strictly a VS Code integration, but it complements it: you can use VS Code for active coding sessions and Telegram for remote tasking.

Deploying on Hostinger

Hostinger offers a one click Docker deploy for Hermes. After deployment, navigate to Docker Manager in the Hostinger panel and click Open to reach the Hermes setup wizard directly inside the container. This wizard will guide you through configuring your model provider and Telegram token.

Setting Up Telegram

1. Create a Telegram bot via BotFather. BotFather will give you a token.
2. Find your personal Telegram user ID using a userinfo bot (like @userinfobot).
3. Lock the bot to your user ID only. This is a security measure so that no other Telegram user can talk to your agent.
4. Provide the token and user ID to Hermes during setup.

Authenticating GitHub CLI

Ask Hermes in the web UI chat to "set up your GitHub CLI skill." Hermes will guide you through the OAuth flow and configure git name and email automatically. The ZazenCodes video recommends using a separate GitHub account or fine grained personal access token for the agent rather than handing over full account credentials. This scopes access to specific repositories.

The Three Layer Architecture in Practice

• VPS: Always on, so Hermes keeps working without your laptop open.
• Docker container: Runs the Hermes image. The /data folder inside the container is mounted as a Docker volume to the host. This folder contains authentication context, skills, and memories.
• Telegram bot: The command interface. You dictate tasks from your phone (e.g., "build a blog, deploy to staging") and Hermes executes the full workflow: loading relevant skills, reading source files, writing code, generating images with Codex, pushing to GitHub, and monitoring the GitHub Action that deploys the site.

Why Transparency in Tool Calls Matters

ZazenCodes explicitly valued being able to see every tool call inside the Telegram chat: which skill is loaded, what terminal command is running, what file is being read. This visibility lets you interrupt, redirect, or approve actions in real time. Knowing exactly what the model is doing, rather than just receiving a final answer, is what makes this workflow feel trustworthy for production code.

Troubleshooting

VS Code Cannot Connect

If VS Code cannot connect to Hermes, run the health check in a terminal:

hermes acp --check

Confirm that the hermes command is on the PATH that VS Code uses. VS Code may use a different PATH than your terminal, especially if you installed Hermes via a package manager that modifies PATH in shell config files. You can check by opening the VS Code terminal (Ctrl+) and running which hermesorhermes --version. If the command is not found, you may need to add the directory containing hermesto VS Code'sterminal.integrated.env` settings.

Browser Tools Missing

If browser tools are missing in ACP mode, run:

hermes acp --setup-browser

Then reconnect the VS Code ACP Client. This installs the necessary dependencies (like Playwright or Puppeteer) that Hermes uses for browser automation.

Hermes Edits the Wrong Files

Open VS Code at the repository root and start the ACP session from that workspace. Hermes uses the workspace root as its working directory. If you open VS Code in a subfolder, Hermes may not see the full project structure.

Terminal Commands Run in the Wrong Environment

Check the active Hermes profile, shell PATH, virtualenv, and workspace path before changing prompts. Hermes runs commands in its own shell environment, which may not match your interactive shell. If you use virtual environments (e.g., Python venv), make sure the environment is activated in the Hermes profile or specify the full path to the interpreter.

Expected Inline Autocomplete

If you expected inline autocomplete, keep Copilot for completions and use Hermes for multi step tool execution, memory, tests, and repo automation. Hermes is not an autocomplete product; it is an editor surface for an autonomous tool using agent.

Telegram Bot Messages Not Arriving

If Telegram messages are not coming through, tell Hermes in the web UI chat that Telegram is not working. It will load its self knowledge skill, check the gateway process, and restart it. The ZazenCodes video demonstrates this exact scenario live. This is a community reported fix.

Hermes Cannot Find the Cloned Repository

Make sure the repository is cloned into the expected path inside the Docker container. Start a new session and ask Hermes to search for the project. It may take a few searches the first time but will learn the path for future sessions through its memory.

Approval Required for a Git Push Command

Hermes flags high risk commands before executing. When prompted in Telegram, simply approve the command. You can approve once or for the session depending on how much autonomy you want. This is a safety feature, not a bug.

Going Further

Once you have the basic VS Code integration working, consider these next steps from the source material:

• Use Hermes with OpenRouter: Set up hosted model access with credits, rate limits, and provider fallback for tool heavy editor sessions. This is useful if you want to use multiple models or need a fallback when your primary provider is down.
• Connect Hermes to GitHub: Move from local VS Code steering to PR review, CI summaries, issue triage, and repository automation. This works even when the editor is closed.
• Create reusable Hermes skills: Save repo conventions and repeated debugging workflows so future VS Code sessions start smarter. Skills are stored in the /data folder and can be shared across sessions.
• Compare Hermes with GitHub Copilot: Understand when to use inline completions versus an autonomous tool using coding agent. The official documentation provides a clear comparison.
• Deploy on a VPS for 24/7 operation: Use Hostinger's one click Docker deploy to run Hermes on a VPS. This enables the Telegram workflow and lets you code from your phone.
• Set up a dedicated GitHub account for the agent: The ZazenCodes presenter recommends this so the agent's access stays scoped to specific repositories and does not touch everything on your personal account.

For a complete walkthrough of the remote VPS + Telegram workflow, watch the ZazenCodes video linked in the source material. It covers the full stack from deployment to live demo.