Hermes Agent Dashboard
Dashboard is Hermes's web-based visual management interface.
Through the browser, you can centrally manage all Profile configurations, models, skills, MCP servers, scheduled tasks, and sessions — without needing to remember any command-line arguments.
This chapter introduces each functional module of the Dashboard one by one, and how to use it to improve daily operations efficiency.
What is the Dashboard?
If the command line is like Hermes' terminal interface, the Dashboard is its control center.
Dashboard is a locally running web application, accessed through a browser, providing a graphical management experience.
Its core value lies in:
- Unified management of multiple profilesSwitching and managing all Profiles in the same interface, without needing to repeatedly launch different terminal windows
- Visualization configurationDrop-down menus to select models, drag-and-drop to enable skills, one-click MCP server installation, lowering the configuration threshold.
- Usage at a glanceToken consumption, cost statistics, and session history displayed in chart form.
- Online chatChat directly with any Profile's Agent in the browser to test configuration effects.
The Dashboard is not a replacement for the command line, but a complementary tool. The command line suits quick operations and scripting; the Dashboard suits browsing, exploring, and managing.
Launch and access
Basic startup
Start the Dashboard with a single command:
hermes dashboard
After startup, open it in the browserhttp://localhost:9119Then you can access it.
Startup options
# 启动并预选指定 Profile（跳过 Profile 选择步骤） coder dashboard # 指定自定义端口 hermes dashboard --port 8080 # 绑定到所有网络接口（谨慎使用，建议配合防火墙） hermes dashboard --port 8080 --host 0.0.0.0
| Options | Description | Default value |
|---|---|---|
| --port | Web service listening port | 9119 |
| --host | Bound network interface address | 127.0.0.1 (local only) |
By default, the Dashboard only binds to 127.0.0.1, accessible only from the local machine. If you need to access it from other devices (such as remote management after deploying on a server), use --host 0.0.0.0 and combine it with firewall rules to restrict source IPs.
Launch in Docker environment
In Docker deployment, Dashboard runs as an independent service:
Example
# Dashboard as a standalone service
services:
dashboard:
image: nousresearch/hermes-agent:latest
command: hermes dashboard --port 9119
network_mode: host
volumes:
- ~/.hermes:/home/hermes/.hermes
ports:
- "127.0.0.1:9119:9119"
restart: unless-stopped
Functional module details
Profile switcher
At the top of the sidebar is the Profile switcher, which is Dashboard's most important navigation component.
Its core capability is to switch instantly between different Profiles without restarting Dashboard:
- The dropdown menu lists all created profiles (default, coder, researcher, etc.)
- After switching, all modules below—Models, Skills, MCP, Chat—automatically refresh to show the current Profile's data
- The currently selected Profile name is always displayed at the top, preventing accidental misoperation
This solves the most common pain point in multi-Profile management: previously you had to remember "what model coder uses, what skills researcher has installed," now with a switch you can see everything.
Models Management
The Models module handles model configuration and usage monitoring, and is one of the most frequently used features in the Dashboard.
Main model configurationSelect the default model from the drop-down menu, with filtering by provider (OpenAI, Anthropic, OpenRouter, etc.).
After switching models, the configuration takes effect immediately—the current session uses the old model, new sessions use the new model.
Auxiliary Model ConfigurationAssign independent models to specific tasks to reduce costs:
| Auxiliary Tasks | Purpose | Recommended Models |
|---|---|---|
| background_review | Background self-improvement review | Low-cost models (such as gemini-flash) |
| vision | Image analysis and description | Vision-capable models |
| session_search | Cross-session search summaries | Low-cost Models |
Usage and cost statisticsThe Models page displays in chart form:
- Cumulative token consumption per model (input + output)
- Cumulative cost estimated in USD
- Recent usage trends (past 7 days / 30 days)
- Usage comparison grouped by Profile
Auxiliary models are the most effective way to control costs. Take background_review as an example — each review with claude-sonnet might cost $0.02, while switching to gemini-flash might cost only $0.001. Over dozens of conversations per day, the difference is significant.
Skills Management
The Skills module provides a complete skill management interface, covering the full lifecycle from browsing to installing to creating.
Skill browsing and search：
- Grid view displays all installed skills, with each card showing the name, description, and category tags.
- The search box supports filtering by name and description keywords
- Browse grouped by category directories (software-development, research, devops, etc.)
Skill Installation：
- One-click installation of built-in optional skills
- Search and install community skills from the agentskills.io Hub
- Install skills directly from a URL (paste the public link to SKILL.md).
- After installation, it automatically appears in the slash command list.
Learn a skill panelThis is a Dashboard-exclusive feature — graphical skill creation./learnCommand palette.
It provides three input boxes, corresponding to three learning sources:
| input box | Input Content | Example |
|---|---|---|
| Directory | Local documentation or project path | ~/projects/acme-sdk/docs |
| URL | Online documentation or API reference URL | https://docs.example.com/api |
| Free Text | Describe workflow or paste notes | "Reimbursement process: open the portal, create a new expense report..." |
After filling in any one or more input boxes, click the Learn button, and the Agent will collect materials and generate a SKILL.md draft, which you preview and then confirm to save.
MCP Server Management
The MCP module transforms command-linehermes mcpoperations into a visual interface.
MCP Directory Browsing：
- Grid view displays all entries in the Nous curated catalog
- Each entry displays name, description, and current status (available / enabled / disabled)
- Status labels are color-coded—green enabled, gray available, yellow disabled
Installation and Configuration：
- Click an entry to go to the detail page and view the list of tools exposed by the server.
- One-click installation—automatically guides through API Key configuration or OAuth login
- After installation, you can select the specific tools to enable in the tool selection interface (include/exclude filtering).
Installed Server Management：
- View detailed information about all installed servers.
- Reconfigure the tool selection list
- Enable/disable servers without editing config.yaml.
- Test server connection status
Chat Online Chat
The Chat module lets you talk directly with the current Profile's Agent in the browser.
Its interface resembles a messaging app and supports:
- Text message input with streaming reply display
- Real-time display of tool call process (you can see which tool the Agent is calling, along with the parameters and results)
- Message history is retained within the current session (not retained after closing the browser tab).
- Supports all slash commands (/new, /skills, /memory, etc.)
The main purpose of the Chat module isQuickly Test Configuration Effect—after modifying model or skill configurations, you can send a message directly in Dashboard to verify without switching to the terminal.
Chat module sessions are temporary — conversation history is not retained after closing the tab. If you need a persistent chat experience, use the CLI (hermes chat) or messaging platforms (Telegram/Discord).
Cron Scheduled Task Management
The Cron module provides visual creation and management of scheduled tasks.
Task List：
- A table displays all scheduled tasks, with columns including: name, schedule, target platform, status (enabled/paused).
- Supports filtering by status (All / Running / Paused)
- Beside each task there are quick action buttons: run once now, pause, resume, delete.
Create taskA graphical task creation form, with fields including:
- Task name (unique identifier)
- Schedule — supports natural language input (e.g., "every day at 8am") and cron expressions.
- Prompt content — the task description that the Agent executes each time.
- Target platform - where task results are delivered
- Target chat ID — which specific conversation to deliver to.
- Enabled tool set — the tools available when the task is executed.
Execution HistoryView the trigger time, completion status, and brief result summary of each execution.
Other Features
Configuration ManagementSome Dashboard pages allow direct editing of common settings in config.yaml, including:
- Terminal backend switching (local / docker / ssh, etc.)
- Approval mode settings (manual / smart / off)
- Display preferences (such as the detail level of memory notifications).
Checkpoint managementView and restore filesystem checkpoints, same as the command line.hermes checkpoints listThe functions are identical.
Plugin ManagementView installed plugins, enable or disable plugins.
Dashboard Best Practices
Daily usage scenarios
| Scenarios | Recommended practice |
|---|---|
| New Profile Initialization | Use Dashboard's Profile switcher to create a new Profile, then select models on the Models page, skills on the Skills page, and add servers on the MCP page—the entire process is visual, so no steps are missed |
| Model switching evaluation | After changing the model, test with the same prompt on the Chat page to compare response quality and speed across different models. |
| Cost Monitoring | Open the Dashboard's Models page weekly to review usage trends and cost statistics, catching abnormal consumption early. |
| Skill Discovery | Browse the Hub catalog on the Skills page, one-click install interesting skills you come across, much more efficient than command-linehermes skills searchMuch More Intuitive |
| Scheduled task debugging | After creating a Cron task, test it once with the "Run Now" button, and only enable automatic scheduling after confirming the output meets expectations. |
| Multi-agent operations | Switch between different Profiles in Dashboard to view their respective gateway status, Token consumption, and scheduled tasks—manage everything from one interface |
Security Recommendations
- The Dashboard binds to 127.0.0.1 by default; if remote access is needed, be sure to configure firewall rules to restrict source IPs.
- Do not expose the Dashboard port on untrusted networks.
- When deploying with Docker, the Dashboard service should be placed in the internal network (accessible only internally), and accessed from outside via the Gateway's port mapping.
Quick command reference
# ─── 启动 Dashboard ───────────────────────────────────────── hermes dashboard # 默认端口 9119，仅本机访问 hermes dashboard --port 8080 # 指定端口 coder dashboard # 启动并预选 coder Profile # ─── Docker 中运行 ────────────────────────────────────────── docker run -d \ --name hermes-dashboard \ -v ~/.hermes:/home/hermes/.hermes \ -p 127.0.0.1:9119:9119 \ nousresearch/hermes-agent:latest \ hermes dashboard # ─── Dashboard 中可执行的操作 ────────────────────────────── # Models 页：选择主模型、配置辅助模型、查看用量统计 # Skills 页：浏览、搜索、安装技能，使用 Learn a skill 面板 # MCP 页：浏览目录、一键安装、管理工具过滤 # Chat 页：在线与 Agent 对话，测试配置效果 # Cron 页：创建、暂停、恢复定时任务，查看执行历史other extensions