Hermes Setup

Hermes is a community-maintained, open-source AI Agent CLI (hermes-cli) — completely free, with no cloud dependency. It ships no model of its own; instead it acts as a local Agent scheduler that talks to any OpenAI-compatible service — Ollama / LM Studio / vLLM.

Flowix Memo treats Hermes as a plug-in Agent. Once a local model backend is connected, you can invoke the local model directly from the document cursor position — no terminal switching, no copy-pasting context, no manual conversation history management. Hermes is complementary to Claude Code / Codex — slightly less capable, with speed depending on hardware — but data never leaves your machine and it costs nothing. It fits scenarios where corporate policy forbids data egress, disconnected environments, or zero-cost model runs. A single document can host multiple Hermes cards, each running its own independent conversation stream.

1. Install Hermes

Prerequisites:

• You need Python 3.10 or newer
• macOS users should use Homebrew (brew install python) or pyenv. The direct installer is not recommended (you may hit pip permission issues later)
• Windows users also need Git for Windows
• Running local models requires enough memory: 8GB+ for 7B models, 32GB+ for 32B models

Open a terminal and install Hermes:

pip install hermes-cli

Upgrade to the latest version:

pip install --upgrade hermes-cli

Note: macOS users who hit pip permission issues during install should install Python via Homebrew / pyenv. After a successful install, you still need to complete "2. Using Hermes in Flowix Memo" below to finish the connection — do not run the hermes command directly yet (it will not respond before a local model backend is running).

1.1 macOS

Step 1: Install Python 3.10+

brew install python

Step 2: Install the Hermes CLI globally

pip install hermes-cli

1. Verify the install: hermes --version — a version output means the CLI is OK
2. Upgrade (optional): pip install --upgrade hermes-cli
3. End-to-end check: run hermes --help in the terminal — version and help output means the CLI is OK; a real "hello" test waits until section 1.2 brings up the local model backend.

1.2 Windows 11

1. Install Python 3.10+: grab the LTS version from python.org and tick Add to PATH during install
2. Install Git for Windows
3. Install the CLI globally: run pip install hermes-cli in an Administrator PowerShell
4. Verify the install: hermes --version — a version output means the CLI is OK
5. Upgrade (optional): pip install --upgrade hermes-cli
6. End-to-end check: open a fresh terminal and run hermes --help — if it runs, both the CLI and PATH are wired up; a real "hello" test waits until "2" below brings up the local model backend.

2. Using Hermes in Flowix Memo

Once the CLI is installed, Flowix Memo auto-detects it. But Hermes ships no model, so there is one extra step: start a local model backend first, then let Flowix know the backend address.

Start a local model backend (pick one):

• Ollama (simplest, zero config) — ollama serve (default http://localhost:11434/v1), then ollama pull qwen2.5-coder:7b to pull a model
• LM Studio — load a GGUF model, start the OpenAI Compatible Server in the Developer panel (default http://localhost:1234/v1)
• vLLM (production-grade) — vllm serve Qwen/Qwen2.5-Coder-32B-Instruct --port 8000

Once the backend is up, in Flowix Settings > AI > Hermes fill in:

After saving, Flowix pings /v1/models to verify reachability; a green checkmark appears when reachable.

2.1 Insert an Agent card via command

Type / anywhere in a document to bring up the Agent type menu, pick Hermes, and a draggable Agent card is inserted.

2.2 AI Agent conversation

Invoke the local model directly from the document cursor position — no terminal switching, no copy-pasting context, no manual conversation history management. Response speed depends on local hardware — a 7B model runs fine on 8GB RAM and is quick for light tasks; for long-form writing or multi-file refactors, move up to 32B+ or switch to Claude Code.