root@ailaba:~ 31# curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash ┌─────────────────────────────────────────────────────────┐ │ ⚕ Hermes Agent Installer │ ├─────────────────────────────────────────────────────────┤ │ An open source AI agent by Nous Research. │ └─────────────────────────────────────────────────────────┘ ✓ Detected: linux (ubuntu) → Root install on Linux — using FHS layout → Code: /usr/local/lib/hermes-agent → Command: /usr/local/bin/hermes → Data: /root/.hermes (unchanged) → Checking for uv package manager... ✓ uv found (uv 0.11.14 (x86_64-unknown-linux-gnu)) → Checking Python 3.11... → Python 3.11 not found, installing via uv... Installed Python 3.11.15 in 31.53s + cpython-3.11.15-linux-x86_64-gnu (python3.11) ✓ Python installed: Python 3.11.15 → Checking Git... ✓ Git 2.43.0 found → Checking Node.js (for browser tools)... ✓ Node.js v24.15.0 found → Checking internet connectivity for package install and web tools... ✓ Internet connectivity looks good → Checking ripgrep (fast file search)... → Checking ffmpeg (TTS voice messages)... → Installing ripgrep ffmpeg... Reading package lists... Done Building dependency tree... Done Reading state information... Done The following additional packages will be installed: adwaita-icon-theme alsa-topology-conf alsa-ucm-conf at-spi2-common at-spi2-core dconf-gsettings-backend dconf-service fontconfig gsettings-desktop-schemas gtk-update-icon-cache hicolor-icon-theme humanity-icon-theme i965-va-driver intel-media-va-driver libaacs0 libasound2-data libasound2t64 libass9 libasyncns0 libatk-bridge2.0-0t64 libatk1.0-0t64 libatspi2.0-0t64 libavahi-client3 libavahi-common-data libavahi-common3 libavc1394-0 libavcodec60 libavdevice60 libavfilter9 libavformat60 libavutil58 libbdplus0 libblas3 libbluray2 libbs2b0 libcairo-gobject2 libcairo2 libcdio-cdda2t64 libcdio-paranoia2t64 libcdio19t64 libchromaprint1 libcjson1 libcodec2-1.2 libcolord2 libcups2t64 libdatrie1 libdav1d7 libdc1394-25 libdconf1 libdecor-0-0 libdecor-0-plugin-1-gtk libepoxy0 libfftw3-double3 libflac12t64 libflite1 libgfortran5 libgme0 libgsm1 libgtk-3-0t64 libgtk-3-bin libgtk-3-common libhwy1t64 libiec61883-0 libigdgmm12 libjack-jackd2-0 libjxl0.7 liblapack3 liblilv-0-0 libmbedcrypto7t64 libmp3lame0 libmpg123-0t64 libmysofa1 libnorm1t64 libogg0 libopenal-data libopenal1 libopenmpt0t64 libopus0 libpango-1.0-0 libpangocairo-1.0-0 libpangoft2-1.0-0 libpgm-5.3-0t64 libpixman-1-0 libplacebo338 libpocketsphinx3 libpostproc57 libpulse0 librabbitmq4 librav1e0 libraw1394-11 librist4 librsvg2-2 librsvg2-common librubberband2 libsamplerate0 libsdl2-2.0-0 libserd-0-0 libshine3 libsnappy1v5 libsndfile1 libsndio7.0 libsord-0-0 libsoxr0 libspeex1 libsphinxbase3t64 libsratom-0-0 libsrt1.5-gnutls libssh-gcrypt-4 libsvtav1enc1d1 libswresample4 libswscale7 libthai-data libthai0 libtheora0 libtwolame0 libudfread0 libunibreak5 libva-drm2 libva-x11-2 libva2 libvdpau1 libvidstab1.1 libvorbis0a libvorbisenc2 libvorbisfile3 libvpl2 libvpx9 libwayland-cursor0 libwayland-egl1 libx264-164 libx265-199 libxcb-render0 libxdamage1 libxss1 libxvidcore4 libzimg2 libzix-0-0 libzmq5 libzvbi-common libzvbi0t64 mesa-va-drivers mesa-vdpau-drivers ocl-icd-libopencl1 pocketsphinx-en-us session-migration ubuntu-mono va-driver-all vdpau-driver-all Suggested packages: ffmpeg-doc i965-va-driver-shaders alsa-utils libasound2-plugins libcuda1 libnvcuvid1 libnvidia-encode1 libbluray-bdj colord cups-common libfftw3-bin libfftw3-dev gvfs jackd2 libportaudio2 opus-tools pulseaudio libraw1394-doc librsvg2-bin serdi sndiod sordi speex opencl-icd libvdpau-va-gl1 The following NEW packages will be installed: adwaita-icon-theme alsa-topology-conf alsa-ucm-conf at-spi2-common at-spi2-core dconf-gsettings-backend dconf-service ffmpeg fontconfig gsettings-desktop-schemas gtk-update-icon-cache hicolor-icon-theme humanity-icon-theme i965-va-driver intel-media-va-driver libaacs0 libasound2-data libasound2t64 libass9 libasyncns0 libatk-bridge2.0-0t64 libatk1.0-0t64 libatspi2.0-0t64 libavahi-client3 libavahi-common-data libavahi-common3 libavc1394-0 libavcodec60 libavdevice60 libavfilter9 libavformat60 libavutil58 libbdplus0 libblas3 libbluray2 libbs2b0 libcairo-gobject2 libcairo2 libcdio-cdda2t64 libcdio-paranoia2t64 libcdio19t64 libchromaprint1 libcjson1 libcodec2-1.2 libcolord2 libcups2t64 libdatrie1 libdav1d7 libdc1394-25 libdconf1 libdecor-0-0 libdecor-0-plugin-1-gtk libepoxy0 libfftw3-double3 libflac12t64 libflite1 libgfortran5 libgme0 libgsm1 libgtk-3-0t64 libgtk-3-bin libgtk-3-common libhwy1t64 libiec61883-0 libigdgmm12 libjack-jackd2-0 libjxl0.7 liblapack3 liblilv-0-0 libmbedcrypto7t64 libmp3lame0 libmpg123-0t64 libmysofa1 libnorm1t64 libogg0 libopenal-data libopenal1 libopenmpt0t64 libopus0 libpango-1.0-0 libpangocairo-1.0-0 libpangoft2-1.0-0 libpgm-5.3-0t64 libpixman-1-0 libplacebo338 libpocketsphinx3 libpostproc57 libpulse0 librabbitmq4 librav1e0 libraw1394-11 librist4 librsvg2-2 librsvg2-common librubberband2 libsamplerate0 libsdl2-2.0-0 libserd-0-0 libshine3 libsnappy1v5 libsndfile1 libsndio7.0 libsord-0-0 libsoxr0 libspeex1 libsphinxbase3t64 libsratom-0-0 libsrt1.5-gnutls libssh-gcrypt-4 libsvtav1enc1d1 libswresample4 libswscale7 libthai-data libthai0 libtheora0 libtwolame0 libudfread0 libunibreak5 libva-drm2 libva-x11-2 libva2 libvdpau1 libvidstab1.1 libvorbis0a libvorbisenc2 libvorbisfile3 libvpl2 libvpx9 libwayland-cursor0 libwayland-egl1 libx264-164 libx265-199 libxcb-render0 libxdamage1 libxss1 libxvidcore4 libzimg2 libzix-0-0 libzmq5 libzvbi-common libzvbi0t64 mesa-va-drivers mesa-vdpau-drivers ocl-icd-libopencl1 pocketsphinx-en-us ripgrep session-migration ubuntu-mono va-driver-all vdpau-driver-all 0 upgraded, 150 newly installed, 0 to remove and 31 not upgraded. Need to get 108 MB of archives. After this operation, 306 MB of additional disk space will be used. Get:1 http://archive.ubuntu.com/ubuntu noble-updates/main amd64 gtk-update-icon-cache amd64 3.24.41-4ubuntu1.3 [51.9 kB] Get:2 http://archive.ubuntu.com/ubuntu noble/main amd64 hicolor-icon-theme all 0.17-2 [9976 B] Get:3 http://archive.ubuntu.com/ubuntu noble/main amd64 humanity-icon-theme all 0.6.16 [1282 kB] Get:4 http://archive.ubuntu ...... ## 2353 how hermes do web search? --- 1. How Hermes Actually Searches (The Default Way) When you ask Hermes to search the web, it usually executes a two-step process using Python libraries under the hood: Step 1: The Search API: It doesn't physically "Google" things. It pings a dedicated Search API (like DuckDuckGo, Tavily, or Brave Search). This returns a clean JSON list of URLs, titles, and short snippets. Step 2: The Scraper: If it needs to read a specific page, it uses standard HTTP libraries (like requests or aiohttp) to download the raw HTML. It then passes that HTML through a parser (like BeautifulSoup or Mozilla's Readability) to strip away all the CSS, JavaScript, and navigation menus, leaving only the core text. Why it avoids TUI browsers: Tools like w3m and lynx format text specifically for human eyeballs on a 2D grid. They add line breaks, spaces, and visual indentations that actually waste an LLM's precious context window. Pure, stripped markdown is much more token-efficient. --- You are exactly right to be suspicious. Out of the box, Hermes does not use SearXNG; it defaults to **Firecrawl** for all web interactions. However, a recent architectural update to Hermes-Agent explicitly split the web tools into different phases, allowing you to plug **SearXNG** in as a native backend. Here is exactly how to configure it inside your **ailaba** container. ### 1. Edit the Config File Hermes stores its core behavioral settings in a YAML file. Open it with your editor: ```bash vim ~/.hermes/config.yaml ``` You need to find (or add) the `web` block and explicitly set the `search_backend` to SearXNG: ```yaml web: search_backend: searxng # Keep an extract backend here, explained below extract_backend: firecrawl ``` ### 2. Point to Your SearXNG Instance Hermes needs to know where your SearXNG server lives. You configure this in the environment file. Open your secrets/env file: ```bash vim ~/.hermes/.env ``` Add the URL pointing to your instance: ```bash SEARXNG_URL="http://your-searxng-ip:port" ``` *Crucial Server Setup Note:* If you are self-hosting SearXNG, you must ensure that the `json` format is enabled in its `settings.yml` (it is sometimes disabled by default for privacy). If Hermes can't get a JSON response, it will crash when trying to read the search results. --- ### 3. The "Search" vs. "Extract" Catch Because you are customizing the stack, you need to understand how Hermes divides the labor: * **The Search Phase (`web_search`):** This takes a query and returns a list of URLs and short text snippets. **SearXNG** is perfect for this, as it acts as a massive metasearch engine. * **The Extract Phase (`web_extract`):** SearXNG *cannot* extract full-page content. If Hermes decides a snippet isn't enough and wants to read the entire 5,000-word article, it needs an extraction backend. If you completely remove Firecrawl from the `extract_backend` config, Hermes will fall back to making a plain HTTP request. This works for simple blogs, but it will instantly fail on any modern, JavaScript-rendered website or anything sitting behind Cloudflare/bot protection. If you want to keep everything strictly local and free on your **zaasus** machine, you can leave SearXNG as the searcher, and instruct Hermes to use its built-in local Chromium browser tools to navigate to and snapshot the pages, or even use that `w3m -dump` trick we discussed earlier to read the raw text! Do you already have a SearXNG Docker container running on your host bridge (`10.12.88.1`), or are you planning to use a public SearXNG instance for the agent? # May 14 ## 0138 You are running into a classic "greedy framework" issue. When tools like Hermes boot up, they often scan the environment first. If they spot an `OPENAI_API_KEY` or `ANTHROPIC_API_KEY` in your `.env` file, they aggressively assume you want to use the cloud, overriding your local settings. To force Hermes to ignore the cloud keys and strictly use your Ollama instance at `10.12.88.1:11434`, you need to lock down the **chat configuration**. Here is how to do it permanently inside your **ailaba** container. ### 1. The CLI Method (Recommended) Since we know Hermes likes to overwrite the YAML file, the safest way is to use the built-in configuration commands to enforce the custom OpenAI-compatible endpoint. Run these three commands in your terminal: ```bash # 1. Tell Hermes to use the standard OpenAI protocol (which Ollama supports natively) hermes config set chat.provider openai # 2. Point it strictly to your host's bridge IP hermes config set chat.base_url "http://10.12.88.1:11434/v1" # 3. Set your specific local model (e.g., the 16k one we built earlier) hermes config set chat.model "hermes-16k" ``` ### 2. The YAML Verification If you want to open your `config.yaml` to verify it worked (or just prefer editing it with Vim), it should look exactly like this: ```bash vim ~/.hermes/config.yaml ``` Look for the `chat:` block. It needs these specific lines: ```yaml chat: provider: "openai" model: "hermes-16k" base_url: "http://10.12.88.1:11434/v1" ``` *(Note: Even though the provider says "openai", the `base_url` hijacks the request and sends it straight to your local RTX 4050 instead).* ### 3. The "Dummy Key" Fallback Because we set the provider to "openai", some underlying Python libraries inside Hermes might still throw a tantrum if they don't find *any* OpenAI key in the environment, even though Ollama doesn't need one. To prevent this without accidentally connecting to the real OpenAI, give it a fake key in your environment file: ```bash vim ~/.hermes/.env ``` Add this line: ```bash OPENAI_API_KEY="ollama" ``` Once those three settings are locked in, Hermes will stop reaching out to the cloud and will route all its thinking strictly through your host machine's GPU! # May 14 ## 1251 Hermes has no /ask /chat /talk, so how to simple /ask the model ? My hermes currently doesn't show model name, which should be ollama/gemma4:e2b . no ctx number too, how to make up these 2 settings? ``` 36# hermes ╔══════════════════════════════════════════════════════════════════════════╗ ║ ⚕ NOUS HERMES - AI Agent Framework ║ ║ Hermes Agent v0.13.0 (2026.5.7) · upstream 942adf61 ║ ╚══════════════════════════════════════════════════════════════════════════╝ ● · 30 tools · toolsets: browser, clarify, code_execution, computer_use, cronjob, delegation, file, image_gen, kanban, memory, messaging, session_search, skills, terminal, todo, tts, vision, web · provider: auto Welcome to Hermes Agent! Type your message or /help for commands. ✦ Tip: hermes status --deep runs the full health sweep across every component; plain hermes status is the quick view. ⚕ unknown │ ctx -- │ [░░░░░░░░░░] -- │ 44s │ ⏲ 0s ──────────────────────────────────────────────────────────────────────────── ❯ ──────────────────────────────────────────────────────────────────────────── ``` ## 1310 I don't think hermes is being hijacked to local bridge :11434, when i use /model , it can pickup keys from env. and the settings in config.yaml just not work for it. And i guess the web search is from google also, not a locally initialized websearch. ``` ✓ Model switched: gemini-2.5-flash-lite Provider: Google Context: 1,048,576 tokens Max output: 65,536 tokens Cost: $0.10/M in, $0.40/M out, cache read $0.01/M Capabilities: reasoning, tools, vision, PDF, audio, structured output (session only — add --global to persist) ──────────────────────────────────────── ● where is the city of NewYork? Initializing agent... ──────────────────────────────────────── ┊ 🔍 preparing web_search… ┊ 🔍 search where is the city of New York 6.5s ╭─ ⚕ Hermes ───────────────────────────────────────────────────────────────╮ The city of New York is located at the southern tip of New York State in the northeastern United States. It is situated at the mouth of the Hudson River. ╰──────────────────────────────────────────────────────────────────────────╯ ⚕ gemini-2.5-flash-lite │ 15.3K/1M │ [░░░░░░░░░░] 1% │ 16m │ ⏲ 10s ──────────────────────────────────────────────────────────────────────────── ❯ ``` --- To configure [Hermes Agent (by Nous Research)](https://github.com/nousresearch/hermes-agent) to perform web searches, choose between utilizing the zero-key Nous Tool Gateway (Paid Subscription) or integrating a free local SearXNG Docker container. [1, 2, 3, 4] ------------------------------ ## Option 1: Setup via Nous Tool Gateway (Easiest) [3] Paid Nous Portal subscribers can route web_search and web_extract directly through Firecrawl without requiring individual third-party provider API keys. [2] 1. Select the Portal Model: Run the model wizard and choose Nous Portal to automatically engage the gateway: hermes model 2. Toggle the Web Tools: Open your tools manager: hermes tools 3. Select the web search categories and set the provider to Nous Subscription (which updates your config.yml to reflect use_gateway: true). 4. Verify Deployment: hermes status [2, 3, 5] ------------------------------ ## Option 2: Setup via Local SearXNG (Free / Self-Hosted) To avoid subscription fees, self-host a SearXNG instance and configure Hermes to connect to it locally. [4] 1. Launch SearXNG: Build and run the container: docker run -d -p 8888:8080 --name searxng --restart unless-stopped searxng/searxng:latest 2. Enable JSON Output: SearXNG disables JSON format by default, but Hermes requires it. Copy the auto-generated config out of your container: docker cp searxng:/etc/searxng/settings.yml ~/settings.yml 3. Open ~/settings.yml in a text editor, navigate to the formats: block (typically around line 84), and append - json: formats: - html - json 4. Apply and Restart: Copy the edited file back and restart the container: docker cp ~/settings.yml searxng:/etc/searxng/settings.yml docker restart searxng 5. Link Hermes to SearXNG: Point Hermes to your local server instance: hermes config set SEARXNG_BASE_URL "http://localhost:8888/" [4] ------------------------------ ## Option 3: Browser Harness (Full Automation) If your goal is automated browsing (clicking, typing, and logging into websites) instead of raw search strings, configure the browser harness. [6] 1. Install Browser Use Skill: hermes skills install skills-sh/browser-use/browser-use/browser-use 2. Launch with Chromium: Enable remote debugging on your local Google Chrome app, or buy a cloud provider token and configure it: hermes config set BROWSER_USE_API_KEY "your_key_here" [6, 7] ------------------------------ ## Initiating a Search Task Once configured via any of the methods above, initiate the agent through the CLI and explicitly assign a web-based prompt: [3, 6] hermes chat "Search the web for the latest updates on the Hermes Agent v0.13 framework and summarize it." If you ran into any errors during setup, let me know: * Your current Hermes Agent version (hermes --version) * Whether you prefer a cloud gateway or a completely local setup * The specific LLM provider you are using (OpenRouter, OpenAI, local Ollama, etc.) [1, 6, 8, 9, 10] [1] [https://github.com](https://github.com/nousresearch/hermes-agent) [2] [https://www.firecrawl.dev](https://www.firecrawl.dev/blog/hermes-web-search) [3] [https://qiita.com](https://qiita.com/kai_kou/items/84aee2e626c01e222280) [4] [https://hermes-agent.nousresearch.com](https://hermes-agent.nousresearch.com/docs/user-guide/features/web-search) [5] [https://www.firecrawl.dev](https://www.firecrawl.dev/blog/hermes-web-search) [6] [https://www.youtube.com](https://www.youtube.com/watch?v=ES_gBHPAcJU) [7] [https://docs.browser-use.com](https://docs.browser-use.com/cloud/tutorials/integrations/hermes-agent) [8] [https://docs.openwebui.com](https://docs.openwebui.com/getting-started/quick-start/connect-an-agent/hermes-agent/) [9] [https://felo.ai](https://felo.ai/blog/hermes-agent-tutorial-research-workflow/) [10] [https://www.reddit.com](https://www.reddit.com/r/AISEOInsider/comments/1t2vvtx/hermes_agent_desktop_app_runs_powerful_ai_without/)