What you need

A machine with Python 3.10+ installed.

An API key from your model provider (OpenAI, Anthropic, or any OpenAI-compatible endpoint).

Hermes Agent installed via pip: `pip install hermes-agent`.

Step 1: Understand the routing concept

Hermes Agent uses a router that examines each incoming task and decides which specialist agent should handle it.

Each specialist agent has a defined role, tools, and output format. The router passes context including previous results.

This is different from CrewAI's role-playing or LangGraph's graph model — Hermes optimizes for sequential task routing with context passing.

Step 2: Define your workflow

A typical business workflow: Research → Draft → Review → Publish.

Create four specialist agents: Researcher, Writer, Editor, Publisher.

Each agent has a clear role description and access to specific tools.

Step 3: Configure routing rules

Routing rules tell Hermes when to pass a task from one agent to the next.

Example: 'After the Researcher outputs a brief with cited sources, route to the Writer with the full brief as context.'

Rules can include condition checks: 'Only route to Publisher if Editor approves the draft.'

Step 4: Set handoff criteria

Define what a successful handoff looks like for each step.

Researcher handoff: must include at least 3 sources with citations.

Writer handoff: draft must be within word count and match brand tone guidelines.

Editor handoff: must pass a quality checklist before publishing.

Step 5: Test the workflow end to end

Run the full workflow with a test brief. Monitor how context flows between agents.

Check that each agent receives the correct context from the previous step.

Verify the routing decisions — did the router correctly choose the right specialist each time?

Step 6: Add human review checkpoints

Insert human-in-the-loop checkpoints at critical decision points.

Example: pause before publishing for a human to approve the final content.

Hermes supports async checkpoints — the workflow pauses and notifies the reviewer via email or dashboard.

FAQ

Can Hermes Agent run multiple workflows in parallel?

Yes. Each workflow instance is independent. You can run dozens of parallel workflows, each with its own routing rules and agent assignments.

How does Hermes handle errors in a step?

Hermes supports retry logic with configurable attempts and fallback agents. If a specialist fails, the router can assign a different agent or escalate to a human.

Can I use Hermes with local LLMs?

Yes. Hermes supports any OpenAI-compatible API endpoint, including Ollama, vLLM, and LocalAI.

Build a Hermes Workflow

  1. Install Hermes Agent: pip install hermes-agent
  2. Define specialist agents with roles and tools
  3. Configure routing rules for task flow
  4. Set handoff criteria per step
  5. Insert human review checkpoints
  6. Run and monitor the end-to-end workflow

Products that work with this harness

These business kits are designed to install and run on Hermes Agent. Each includes pre-configured agent roles, prompts, workflows, and approval gates.

Ready to see the full catalog?

Browse all 42 agent-ready business kits. Each includes pre-configured agent roles, prompts, workflows, install steps, and approval gates.

Shop all business kits