Building a Local AI Agent Stack with Hermes Agent v0.14 on Debian

I set up a fully local AI-agent environment on Debian using v0.14.0 together with and Qwen models.

What I Installed

The stack now includes:

  • Hermes Agent v0.14.0
  • Ollama local inference server
  • Qwen2.5 32B model
  • Persistent Hermes gateway service
  • OpenAI-compatible proxy support
  • Local coding/automation agent tools

The final runtime ended up using:

  • Python 3.13.2 via pyenv
  • Debian Linux
  • Conda/Miniforge available alongside Hermes

Challenges Encountered

The installation initially failed because of:

  • Conda environment conflicts
  • Broken uv cache permissions
  • Python interpreter inspection errors

Errors looked like:

Failed to inspect Python interpreter
Permission denied (os error 13)

I fixed this by:

  1. Deactivating Conda temporarily
  2. Cleaning corrupted uv caches
  3. Repairing file ownership with chown
  4. Reinstalling/update paths manually

Key Fixes

I removed broken caches:

sudo rm -rf ~/.cache/uv
sudo rm -rf ~/.local/share/uv

Then repaired ownership:

sudo chown -R $USER:$USER ~/.cache ~/.local/share/uv

And updated Hermes manually:

uv pip install --system --upgrade hermes-agent

Successful Installation

Eventually Hermes launched successfully with:

hermes

The TUI confirmed:

  • 29 tools enabled
  • 100+ skills loaded
  • Browser automation
  • Code execution
  • Computer-use support
  • GitHub tooling
  • MCP integrations
  • Autonomous agent workflows

Gateway + Background Services

Hermes gateway was installed as a persistent user service:

systemctl --user status hermes-gateway

This allowed Hermes to:

  • run continuously in the background,
  • expose messaging integrations,
  • manage automation tasks,
  • and support OpenAI-compatible proxy workflows.

Local Model Integration

Hermes automatically detected the local Ollama server:

Local (127.0.0.1:11434) — qwen2.5:32b

This created a fully local inference stack with no cloud dependency required.

What the Final System Can Do

The completed setup now supports:

  • autonomous coding agents,
  • local LLM inference,
  • browser automation,
  • GitHub repository analysis,
  • Discord/social integrations,
  • scheduled jobs,
  • multi-agent delegation,
  • OpenAI-compatible APIs,
  • and local AI development workflows.

Why Hermes Agent v0.14 Is Interesting

The v0.14 “Foundation Release” significantly improved:

  • Windows/Linux support,
  • PyPI installation,
  • proxy/API compatibility,
  • local model integration,
  • gateway services,
  • and agent orchestration tooling.

Combined with Ollama, it creates a strong self-hosted AI-agent platform for developers experimenting with local-first AI workflows.