Build a Hermes agent
with Fabryka.

Ask your coding agent to configure it.

Copy this prompt into your coding agent from the root of the project you want to connect.

Configure this machine to run Hermes Agent with Fabryka AI as its OpenAI-compatible model provider.

Use:
- Base URL: https://fabryka.ai/v1
- Model: auto
- API key environment variable: FABRYKA_API_KEY

Install Hermes only from its official repository. Ask me for the API key without printing it. Store it in the Hermes environment file with mode 600. Configure streaming and preserve OpenAI-format tool calls. Enable only the minimum toolsets required. Install the background gateway, run one harmless tool-call test, and show me the selected model, resulting service status, and changed files. Do not expose the gateway publicly.

What you need.

• Linux or macOS with Python and terminal access.
• A Fabryka key generated on the homepage.
• A persistent host if the gateway and scheduled tasks must run continuously.

You do not need a local GPU. Inference runs behind https://fabryka.ai/v1.

Install Hermes Agent.

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
hermes --version

Review remote installation scripts before executing them, especially on a shared or production host.

Connect Fabryka.

Run hermes model and select a custom OpenAI-compatible endpoint, or add the provider values to the Hermes environment and configuration files.

FABRYKA_API_KEY=your_key_here
OPENAI_API_KEY=your_key_here
OPENAI_BASE_URL=https://fabryka.ai/v1

model:
  default: auto
  provider: custom
  base_url: https://fabryka.ai/v1

chmod 600 ~/.hermes/.env

Test the endpoint first.

curl https://fabryka.ai/v1/chat/completions \
  -H "Authorization: Bearer $FABRYKA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"auto","messages":[{"role":"user","content":"Reply with: connected"}],"stream":true}'

The response header X-Fabryka-Selected-Model and each completion's model field report the model actually used.

Enable tool calling deliberately.

With model: auto, tool requests are routed to Qwen. The OpenAI tool contract supports tools, tool_choice, assistant tool_calls, and results using role: tool. Hermes executes the selected tool on your host; the model only chooses the function and arguments. Pin qwen3.8-27b only when deterministic Qwen selection matters.

hermes tools

Install the gateway.

hermes gateway install
hermes gateway status
hermes cron status

Keep terminal-capable gateways private or loopback-bound. Use an authenticated tunnel for remote access.

Common failures.